diff --git a/records/sd1/sweep/42ab48d701042f8e90ef/output.webp b/records/sd1/sweep/42ab48d701042f8e90ef/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..cd5f0072470da29944d6ac786029eaf185387be2
--- /dev/null
+++ b/records/sd1/sweep/42ab48d701042f8e90ef/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bfeea5fc1da4e5819b8cb7a1018c099f364a1c1ebe30afb31d4ed29d342f836
+size 19000
diff --git a/records/sd1/sweep/42d10216df763d77649d/output.webp b/records/sd1/sweep/42d10216df763d77649d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..56aa5cb1117e197ebe9ff92da7d3ef22269bb62b
--- /dev/null
+++ b/records/sd1/sweep/42d10216df763d77649d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6fe7300b1e87d9f5fcea7e181974324864b856f1d3060764b38815e55e60418
+size 57732
diff --git a/records/sd1/sweep/42d1eac395c8cc640e59/output.webp b/records/sd1/sweep/42d1eac395c8cc640e59/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a832a6de0e2e9ded01d74746e646e5a593ac1e61
--- /dev/null
+++ b/records/sd1/sweep/42d1eac395c8cc640e59/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81d0b7f0a9cc081b5b3c4aa5c68497ef1ac8077c628f2dc7c17d6dfb08d5b483
+size 17650
diff --git a/records/sd1/sweep/42dc32c95857841c62e4/output.webp b/records/sd1/sweep/42dc32c95857841c62e4/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..272dee30e2ad115e1fed9149e987d62bcdcc7cbf
--- /dev/null
+++ b/records/sd1/sweep/42dc32c95857841c62e4/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb422b19150b3668e80ddab5b4a5ca6f03f4302810222a79dea4f0d475a804cc
+size 20734
diff --git a/records/sd1/sweep/42e5a11aa88946d08e23/output.webp b/records/sd1/sweep/42e5a11aa88946d08e23/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..b5ca95e41734abee50449751f72d937a7e819f4a
--- /dev/null
+++ b/records/sd1/sweep/42e5a11aa88946d08e23/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b22e00cafcc6b9212a1ed6a78bf5303a55b4edf4095202fd8ca6d4588e4c3c4
+size 29132
diff --git a/records/sd1/sweep/42f68ecac5588d2e096d/output.webp b/records/sd1/sweep/42f68ecac5588d2e096d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..161b2a9376666a57b21833a3024216dc019e0175
--- /dev/null
+++ b/records/sd1/sweep/42f68ecac5588d2e096d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:570ae2c1e900c6e61bd2beb28c07abc2b3df1d874b1e50119a7b2cf35ecd1f1b
+size 56264
diff --git a/records/sd1/sweep/42f6f7da4ade21a53aca/output.webp b/records/sd1/sweep/42f6f7da4ade21a53aca/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0322ec67e522071430cffdda3ae0aab5ad13cbf4
--- /dev/null
+++ b/records/sd1/sweep/42f6f7da4ade21a53aca/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13ff3f2064eabad49adf045e85af13ab16e5557413978cf370288d2ff955f5ba
+size 35598
diff --git a/records/sd1/sweep/430b1ad874c5c469e6a3/output.webp b/records/sd1/sweep/430b1ad874c5c469e6a3/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..17014429206f03df6b48e4020437bdde67348e74
--- /dev/null
+++ b/records/sd1/sweep/430b1ad874c5c469e6a3/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3450b604705e8999b6091c3861d4201e433238a6bfb08da454556d0f1c1a023
+size 11780
diff --git a/records/sd1/sweep/43170ffda2573cdfea00/output.webp b/records/sd1/sweep/43170ffda2573cdfea00/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..2c9f86e03c0b8aa790efd052f3e5198893ecbc02
--- /dev/null
+++ b/records/sd1/sweep/43170ffda2573cdfea00/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a7730e4c9abd129c7f6390610ca5e2166b86482a58ad97f0200e2d2cdc58bbf
+size 30680
diff --git a/records/sd1/sweep/4319ec37a3d97cd19703/output.webp b/records/sd1/sweep/4319ec37a3d97cd19703/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..721269b155960f1cbf286fbfae2438a39822a70e
--- /dev/null
+++ b/records/sd1/sweep/4319ec37a3d97cd19703/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6a252473dc55cbc8259e7fc6ea2894f1bd10a7bd9e6bd9cb410d9905875b15f
+size 13902
diff --git a/records/sd1/sweep/432ff095fcc788a9a677/output.webp b/records/sd1/sweep/432ff095fcc788a9a677/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..55125ca79e92217c3c92ca6dd9316566848c5563
--- /dev/null
+++ b/records/sd1/sweep/432ff095fcc788a9a677/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54492e6f0870e6958c53edb8c3cdc6ba830c29af998f9655149c81a3567505bb
+size 40834
diff --git a/records/sd1/sweep/43314c2abc59d9a99837/output.webp b/records/sd1/sweep/43314c2abc59d9a99837/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5d7fc9ac1b04c5ba00d57c3a08e841145b1fc65f
--- /dev/null
+++ b/records/sd1/sweep/43314c2abc59d9a99837/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9391261ad7a1fa510b3534e54d314f864e7626ba7f39f7c6b31566981cbfcba7
+size 40838
diff --git a/records/sd1/sweep/4333dd1b0a2a66f9c596/output.webp b/records/sd1/sweep/4333dd1b0a2a66f9c596/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1c5e027bfeb55a504fefbb2c53d35ead76c803ee
--- /dev/null
+++ b/records/sd1/sweep/4333dd1b0a2a66f9c596/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3236c417f9bd548c46eb2d1ea016e82a6c95099550047ee101707787639a844b
+size 57978
diff --git a/records/sd1/sweep/433b7a09310129ce7106/output.webp b/records/sd1/sweep/433b7a09310129ce7106/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..da940ae65983bc91f4f635cb7654ab4c5dbffe86
--- /dev/null
+++ b/records/sd1/sweep/433b7a09310129ce7106/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49b98e9b1f64d9e64966d40f5c2e81fc174a00c1f7f73580bfa4e93eec577b3c
+size 32600
diff --git a/records/sd1/sweep/433c7a5671b2544d8a63/output.webp b/records/sd1/sweep/433c7a5671b2544d8a63/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..011b390fd353f9b78a5b4d4d06505e6ef4ca9894
--- /dev/null
+++ b/records/sd1/sweep/433c7a5671b2544d8a63/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21275ea27e0a24ea0e2b56ead5d32099cd632b84b03ecc7587c11de81d9b78fb
+size 40830
diff --git a/records/sd1/sweep/435a82e5daecf06d43b4/output.webp b/records/sd1/sweep/435a82e5daecf06d43b4/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..7b3d54e8298ed22963c1e0fc92d46d470a9b6453
--- /dev/null
+++ b/records/sd1/sweep/435a82e5daecf06d43b4/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b30ad80cebadff04b87a186092b380366528a1d8cfcd2ccdab602d018722be22
+size 41346
diff --git a/records/sd1/sweep/43648fdabd909ba98f06/output.webp b/records/sd1/sweep/43648fdabd909ba98f06/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ff2d6701b4c89a6388ce4114a01352c580120410
--- /dev/null
+++ b/records/sd1/sweep/43648fdabd909ba98f06/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9444dd2bc0a13606705efd9ff5648138c7b087346874d8bfe77054e9e08f0d8a
+size 40902
diff --git a/records/sd1/sweep/4365356017c04f460077/output.webp b/records/sd1/sweep/4365356017c04f460077/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1b4efb11ca440dcd225eccd470eb2621a60ba9ae
--- /dev/null
+++ b/records/sd1/sweep/4365356017c04f460077/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5bb7a6db1f7b1aed46668aa2f02c6bcbb20b89dee757868e6dbb47caf17dc5e
+size 16128
diff --git a/records/sd1/sweep/4372d9245680abfb632d/output.webp b/records/sd1/sweep/4372d9245680abfb632d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..3dc0bd5915a3a847f38069c4ed240de55ac48d6c
--- /dev/null
+++ b/records/sd1/sweep/4372d9245680abfb632d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9444abca2a4ba27873d199dab5ad18e3e3adcb3f59000dff3f011fb22b2c4936
+size 23290
diff --git a/records/sd1/sweep/4378d65dfd67d3e45c6a/output.webp b/records/sd1/sweep/4378d65dfd67d3e45c6a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..23246378fe7eedd584b93b7762b27854304fb487
--- /dev/null
+++ b/records/sd1/sweep/4378d65dfd67d3e45c6a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3835cf82478565e3c2ee79eb1e3de273a1131535bd9ecbca1042da6e14c3938
+size 17658
diff --git a/records/sd1/sweep/439574a1cdcff2df3fa3/output.webp b/records/sd1/sweep/439574a1cdcff2df3fa3/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e2736b40003ed8acb6ceb5a97ea63162ee11bc3b
--- /dev/null
+++ b/records/sd1/sweep/439574a1cdcff2df3fa3/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26a3b08b3752171faeb2d4c36988e2849e0305c22b400b404dec0f7e30159af5
+size 34716
diff --git a/records/sd1/sweep/439db81cfee83bfb575e/output.webp b/records/sd1/sweep/439db81cfee83bfb575e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..7915ff3dd83c0579baf9f00edf07e865fb4c792d
--- /dev/null
+++ b/records/sd1/sweep/439db81cfee83bfb575e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a00e0ca0087975e49b9c1e907f3eb837406ebcc2f5ad3c0bd0fed557eb743b8b
+size 24010
diff --git a/records/sd1/sweep/439e1e09e7ed6b839a9a/output.webp b/records/sd1/sweep/439e1e09e7ed6b839a9a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e6169eda8598a45ee28fb278be46a1915cb2a05c
--- /dev/null
+++ b/records/sd1/sweep/439e1e09e7ed6b839a9a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2039c73114e00a696270030ee4de3fdb9151c1c4daf8f021897998de98553854
+size 16986
diff --git a/records/sd1/sweep/43a0e5b856e85955f680/output.webp b/records/sd1/sweep/43a0e5b856e85955f680/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..8b4ab8bfabf90ed035b5281129ec62fe42d36f11
--- /dev/null
+++ b/records/sd1/sweep/43a0e5b856e85955f680/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9830501059ed8817a81e7cbc7fd47536d53ace77809021e2711b1aae76aa217
+size 46260
diff --git a/records/sd1/sweep/43a17501ffbd3f55be73/output.webp b/records/sd1/sweep/43a17501ffbd3f55be73/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5275a7ce058da1f3a170fcaa9c278bad70f50b98
--- /dev/null
+++ b/records/sd1/sweep/43a17501ffbd3f55be73/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e33f7dd4b5ac57f7f113358edb3e1d8ad672dc1ff192739404facd814b0103e
+size 33682
diff --git a/records/sd1/sweep/43a4deb39149ed5aa748/output.webp b/records/sd1/sweep/43a4deb39149ed5aa748/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0e336c547304bfe2845e0ac33af5f266aedf2426
--- /dev/null
+++ b/records/sd1/sweep/43a4deb39149ed5aa748/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c239c82c7ce33ac020c9d6daa08d9c86c63f9d6379ca5f5505078c7b30f7f91
+size 46376
diff --git a/records/sd1/sweep/43a6183768388ec09a2a/output.webp b/records/sd1/sweep/43a6183768388ec09a2a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..89f5637b4fe9128cc609bff976f1fef533405d1b
--- /dev/null
+++ b/records/sd1/sweep/43a6183768388ec09a2a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1a07a8c55c0b091219fdccaa67a6ef671459aab187f1a2946528e3eb322e60b
+size 31322
diff --git a/records/sd1/sweep/43b306734edea4eb4c66/output.webp b/records/sd1/sweep/43b306734edea4eb4c66/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a37d9c5eac30de0fa2fec6d18c74d45f150ad190
--- /dev/null
+++ b/records/sd1/sweep/43b306734edea4eb4c66/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b242f997e07df2334b181d0997e4d25a2277a02b42b9a0c2dfee42d20c1d3bcc
+size 58490
diff --git a/records/sd1/sweep/43c2089949abf3c44301/output.webp b/records/sd1/sweep/43c2089949abf3c44301/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..caa3c35b431f9f2614556740bf26a887423b279a
--- /dev/null
+++ b/records/sd1/sweep/43c2089949abf3c44301/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c1762fb1fa5f3b179bf9628bfc1f94bce1d414f2cc4d050467c084ec8e79e62
+size 29602
diff --git a/records/sd1/sweep/43c70567e9742bf3cb3e/output.webp b/records/sd1/sweep/43c70567e9742bf3cb3e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..cfc44f9378dd29bb31c986e1c2120a2ab067b59e
--- /dev/null
+++ b/records/sd1/sweep/43c70567e9742bf3cb3e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f79e3999e03482ad1114850bb3110e06ab0d2352fec92efb33c6e588f70881a
+size 20820
diff --git a/records/sd1/sweep/43d08a599f699ac9ccfa/output.webp b/records/sd1/sweep/43d08a599f699ac9ccfa/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e8efbbe7441119c92667e3e8984c76a31f9f7bb5
--- /dev/null
+++ b/records/sd1/sweep/43d08a599f699ac9ccfa/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81eca8f43ac1ebc64a6256ba952f4ff3fccbb69e71582fabaaca29a3d67d7aa0
+size 19546
diff --git a/records/sd1/sweep/43d30441ff8209476489/output.webp b/records/sd1/sweep/43d30441ff8209476489/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9e9f4cb469f8cfa1f5e5fa62c3967d4564c141fc
--- /dev/null
+++ b/records/sd1/sweep/43d30441ff8209476489/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d06e287f0e4df56e8eb8c7828de537830d37ff972f81b8263c1007d498b56467
+size 41810
diff --git a/records/sd1/sweep/43d5eee6c28c5d3ef41a/output.webp b/records/sd1/sweep/43d5eee6c28c5d3ef41a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..845c4abada073ccd5d46e51198c11218a3f7828a
--- /dev/null
+++ b/records/sd1/sweep/43d5eee6c28c5d3ef41a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76cee35684f4520bc94d937a467a662aa6da01ca44347d9f6687229fdbe8413e
+size 38466
diff --git a/records/sd1/sweep/43d78d98ecdf07385646/output.webp b/records/sd1/sweep/43d78d98ecdf07385646/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..4a18901d1c3923a235cf37fb72bbac78e27a126c
--- /dev/null
+++ b/records/sd1/sweep/43d78d98ecdf07385646/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b05dbb5a88e187fb5107e3008ff5596a40e0912a58a5e017581d448ae564ebb
+size 23776
diff --git a/records/sd1/sweep/43e19892e40cfe14732a/output.webp b/records/sd1/sweep/43e19892e40cfe14732a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..74dfa68bbe97a18cba6e9bc22b9bc8e2d4afa3fb
--- /dev/null
+++ b/records/sd1/sweep/43e19892e40cfe14732a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5742c22fba9ae570571bc3ba011abec1d04ebf87dd8fd5fdf9ece9b15ca90b35
+size 24686
diff --git a/records/sd1/sweep/43ef6b6770a76d193e16/output.webp b/records/sd1/sweep/43ef6b6770a76d193e16/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5b5db88b6ca8a8c8ae343c91493bdd85f7ca44ff
--- /dev/null
+++ b/records/sd1/sweep/43ef6b6770a76d193e16/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d093dd6d81b188f6e8c711d16ddc955eab61080554cdaea8ee5b4abebc8e840a
+size 42198
diff --git a/records/sd1/sweep/43f6024c48c234d68497/output.webp b/records/sd1/sweep/43f6024c48c234d68497/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..65747f5dba09a04960a7eec161d555d9d16ac252
--- /dev/null
+++ b/records/sd1/sweep/43f6024c48c234d68497/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f6878169992c20d34bef0a89c642fe106ec1644ed99fca68d73ff96753fcf05
+size 19652
diff --git a/records/sd1/sweep/43fee8bb4ca53ec511f7/output.webp b/records/sd1/sweep/43fee8bb4ca53ec511f7/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..6a117e6689e260a1c945144e5bdeb26233406250
--- /dev/null
+++ b/records/sd1/sweep/43fee8bb4ca53ec511f7/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a2ec002e5580f9899a707c58a81d15923352de643e7ff09213f213b58d6b119
+size 27962
diff --git a/records/sd1/sweep/440391adf60ad361dbb7/output.webp b/records/sd1/sweep/440391adf60ad361dbb7/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..30536d97e7c1f5706f71d4043c3245789c042c54
--- /dev/null
+++ b/records/sd1/sweep/440391adf60ad361dbb7/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e61defe2ea66c0134bda69850365359037d7a3d0058bd52454bf831165ba7f8
+size 32564
diff --git a/records/sd1/sweep/44089f937aa830080711/output.webp b/records/sd1/sweep/44089f937aa830080711/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..886da548e7d03e07276d8e1c596bc56cd3ee8c31
--- /dev/null
+++ b/records/sd1/sweep/44089f937aa830080711/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc7faf6d2d23f98d544ff927d6c5319a82b0690828bf22389da21b2b84a5526b
+size 28490
diff --git a/records/sd1/sweep/4421e4399305ef5ebe78/output.webp b/records/sd1/sweep/4421e4399305ef5ebe78/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ed0d8437f3fbe68fb4c783e8407d6873b67448eb
--- /dev/null
+++ b/records/sd1/sweep/4421e4399305ef5ebe78/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae1ad735eac74f19d2cf381bde8e492766eb49747100f28fa467b523b4b698c5
+size 19932
diff --git a/records/sd1/sweep/442870c262150db2fbc2/output.webp b/records/sd1/sweep/442870c262150db2fbc2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..bde3dcad1c2619451d67b04bf5aaa1fb8338a1a8
--- /dev/null
+++ b/records/sd1/sweep/442870c262150db2fbc2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48a7a8f2c97a70f89f5fc1767aaca22a96d72d707c0364639eba248e5e025f72
+size 37178
diff --git a/records/sd1/sweep/4432b4656e54ce657dec/output.webp b/records/sd1/sweep/4432b4656e54ce657dec/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ad3d95375691bd72d21ca059f36976696f0c0e29
--- /dev/null
+++ b/records/sd1/sweep/4432b4656e54ce657dec/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e47f3963524c3fd573260fa1b553dcebac9e36337e32961b5d77534fb51cb48
+size 39122
diff --git a/records/sd1/sweep/44359b99a37cc5188422/output.webp b/records/sd1/sweep/44359b99a37cc5188422/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..664d18721b80a6042cf59e93072a56ed06f6671c
--- /dev/null
+++ b/records/sd1/sweep/44359b99a37cc5188422/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5513040477cfa2ba1caed0c7383c7cc8c95e424a1c6363bd6a7a420c4483e9c8
+size 25318
diff --git a/records/sd1/sweep/4442ddbb1cad049a79c2/output.webp b/records/sd1/sweep/4442ddbb1cad049a79c2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1a53fe32b7734da6a4a3d1a8007a0220e85e9ba3
--- /dev/null
+++ b/records/sd1/sweep/4442ddbb1cad049a79c2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddb6b801c31567f4368f822c31d5f89121d87b040946dc24541d4100a4037844
+size 34916
diff --git a/records/sd1/sweep/44468e653ff7594613dd/output.webp b/records/sd1/sweep/44468e653ff7594613dd/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..eca2e0e6430980584e7c2b32990c00beda6ee298
--- /dev/null
+++ b/records/sd1/sweep/44468e653ff7594613dd/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:932b360519684fb57bdd55770b40df7c503665ace3b83d3e9eaf8ba0b092c546
+size 27272
diff --git a/records/sd1/sweep/4462bf0b8b8c687f3a71/output.webp b/records/sd1/sweep/4462bf0b8b8c687f3a71/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..954a2cf025518fac2711bfa54415273f1469710a
--- /dev/null
+++ b/records/sd1/sweep/4462bf0b8b8c687f3a71/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8dd75b2140462cb5301d3368063c6555aca5aaedd9ce802a2eede0a7f3c3de5a
+size 32072
diff --git a/records/sd1/sweep/446a8627ca633d3c5bbb/output.webp b/records/sd1/sweep/446a8627ca633d3c5bbb/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0b0e90034ec127637c30d17f39ec3a09f95e2ab2
--- /dev/null
+++ b/records/sd1/sweep/446a8627ca633d3c5bbb/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e49a3eb3295a581124613024ff5a84341a25b3f98a7027c516ac587245975a6
+size 34590
diff --git a/records/sd1/sweep/446b337d38db128fa343/output.webp b/records/sd1/sweep/446b337d38db128fa343/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..471211e735f2bc70706449657e242b9859303533
--- /dev/null
+++ b/records/sd1/sweep/446b337d38db128fa343/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:309b1f07db5036578da77f216925a941ace2232bdc9fdddd64164b233962894a
+size 26248
diff --git a/records/sd1/sweep/446cf4f93a1f3c7a6955/output.webp b/records/sd1/sweep/446cf4f93a1f3c7a6955/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..058ba4697e9e8bd5fe7c8928356fc079b07dbde9
--- /dev/null
+++ b/records/sd1/sweep/446cf4f93a1f3c7a6955/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:686cd068726919d000423fabb18f08700d7ee09e99e39d1fabdf96efae1f0a87
+size 21324
diff --git a/records/sd1/sweep/4470fb08bb7682abac1c/output.webp b/records/sd1/sweep/4470fb08bb7682abac1c/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c5b68a67ee14bef0aca2f9c41cdb39dfbebd6753
--- /dev/null
+++ b/records/sd1/sweep/4470fb08bb7682abac1c/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d7f070b370a2e0f412dc86ac896f9f6ea9124ae2513b7b390548cad008f7af8
+size 28192
diff --git a/records/sd1/sweep/44714d8c2cfe1967e007/output.webp b/records/sd1/sweep/44714d8c2cfe1967e007/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9e75e11fa50cb280d9d5da21b5c332bc59e2dcb0
--- /dev/null
+++ b/records/sd1/sweep/44714d8c2cfe1967e007/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6906a97b3f52783feaab68b27a097aa75413b029fcba7efc987ce61a8230100a
+size 15228
diff --git a/records/sd1/sweep/44715c3a8cf01af12d73/output.webp b/records/sd1/sweep/44715c3a8cf01af12d73/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e30859b93a586430dd27a20b962c844d3f36a2c9
--- /dev/null
+++ b/records/sd1/sweep/44715c3a8cf01af12d73/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dad43116d5149987733819907b72f94e673cc1cf5d90f02f166c46702060e8ad
+size 34546
diff --git a/records/sd1/sweep/4471d0340eb332bc485e/output.webp b/records/sd1/sweep/4471d0340eb332bc485e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..98351f5e85f32bf27c6926b112d4c5ca94d2f5bf
--- /dev/null
+++ b/records/sd1/sweep/4471d0340eb332bc485e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a95ca77038202038223029da0cff8647e1e75fa1fc955c912b7ec1bb171c6b8
+size 29354
diff --git a/records/sd1/sweep/4472f9c5b2bdcdc373cf/output.webp b/records/sd1/sweep/4472f9c5b2bdcdc373cf/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..6c115f941171b2f5833ca874e5332a04b5831bc7
--- /dev/null
+++ b/records/sd1/sweep/4472f9c5b2bdcdc373cf/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c00c921eff26676debba38f81ef645c2d475631c8bdc36f9a9c31789a8fdee03
+size 21062
diff --git a/records/sd1/sweep/44780e9e6098e9320492/output.webp b/records/sd1/sweep/44780e9e6098e9320492/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c0c0784d6381265448b7367674475af553853d09
--- /dev/null
+++ b/records/sd1/sweep/44780e9e6098e9320492/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c44fbe8698dc8d1790b6ff440a79e6c3c0ef28848722ca306d2c7bb3257a79c
+size 37412
diff --git a/records/sd1/sweep/447a75c48fd25b7b9e18/output.webp b/records/sd1/sweep/447a75c48fd25b7b9e18/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..04d1e187fc82a2e22562361596b2a2b1b65e1d50
--- /dev/null
+++ b/records/sd1/sweep/447a75c48fd25b7b9e18/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:517e8633a33acb4c09a1121028e638d3f05e0227d63e8f78dd4e5a77ea64b585
+size 38582
diff --git a/records/sd1/sweep/447e600049a41c3e2d83/output.webp b/records/sd1/sweep/447e600049a41c3e2d83/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..8b9938f369777a5834908b6729184e714aeaacde
--- /dev/null
+++ b/records/sd1/sweep/447e600049a41c3e2d83/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9195a18d688b18ac6a2be3172a56b57366cf300854ec85a74eb347685803f716
+size 41152
diff --git a/records/sd1/sweep/44853d561963fe711c3d/output.webp b/records/sd1/sweep/44853d561963fe711c3d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..51e9d79bd6aff05db8bd6ef0ddd1fecb634652c4
--- /dev/null
+++ b/records/sd1/sweep/44853d561963fe711c3d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:093da78fb2cbc8d8f312d5d3117f5aac0f16018b35666e98509323d799685236
+size 23072
diff --git a/records/sd1/sweep/44869224627d11298349/output.webp b/records/sd1/sweep/44869224627d11298349/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a5e92ec315281402d90f45a61927c752c37841c9
--- /dev/null
+++ b/records/sd1/sweep/44869224627d11298349/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9688d7282e0228997d778fb67823d1080790b55348a2af0848fdcf06918213b
+size 19746
diff --git a/records/sd1/sweep/448c3f613d17d6509b45/output.webp b/records/sd1/sweep/448c3f613d17d6509b45/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..2d7051b59567296e13ef7dc38dbc9413e1966258
--- /dev/null
+++ b/records/sd1/sweep/448c3f613d17d6509b45/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d0e13bbe1b02e02c760844ba16985bd0be6acebe33cc1e844604b93f83975a7
+size 102652
diff --git a/records/sd1/sweep/448df51206bb84230487/output.webp b/records/sd1/sweep/448df51206bb84230487/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1286e6f478bca0d4919b15f38ad8cf0b0ac6b57a
--- /dev/null
+++ b/records/sd1/sweep/448df51206bb84230487/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82d9de216aafeb050a81e0e803eb2a57ab37c3e418a23fc424301b2490225b8e
+size 22716
diff --git a/records/sd1/sweep/448f7b36f8835a43b443/output.webp b/records/sd1/sweep/448f7b36f8835a43b443/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a45a707d08e7c77f2d936b3e5219cb6f1fa2a883
--- /dev/null
+++ b/records/sd1/sweep/448f7b36f8835a43b443/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7766b459398250f91fb1c96e1ddfbbbaf9d4f8aee2410ad43514e21c4239eb78
+size 29372
diff --git a/records/sd1/sweep/44901e761fcf228d1133/output.webp b/records/sd1/sweep/44901e761fcf228d1133/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..2bc54c8bcc63492125dfb6833dbf6ed314db520a
--- /dev/null
+++ b/records/sd1/sweep/44901e761fcf228d1133/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c305d29872c577274968f753c933d2c1b0a741759fddea1338844fb4b4416322
+size 51388
diff --git a/records/sd1/sweep/449370344ffa5a5b5076/output.webp b/records/sd1/sweep/449370344ffa5a5b5076/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..bf47f446795bd90e83f70deaffa85fa998a0d0da
--- /dev/null
+++ b/records/sd1/sweep/449370344ffa5a5b5076/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d57489316f4340238df93a08d552781c20b4a66a3f427af90c60e960ba70acd2
+size 16466
diff --git a/records/sd1/sweep/449c49424e9c0d564f23/output.webp b/records/sd1/sweep/449c49424e9c0d564f23/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d51c59b3ab832b36680b452101b36ed7ca2cf254
--- /dev/null
+++ b/records/sd1/sweep/449c49424e9c0d564f23/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5efeb1abfb2909d3da8702add88a7a53333249daaf2d41ab107dae59f933fedd
+size 32394
diff --git a/records/sd1/sweep/44a1dc13c735e657d897/output.webp b/records/sd1/sweep/44a1dc13c735e657d897/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c82bfa859f292515ffd1b890b424578cb8a00ff0
--- /dev/null
+++ b/records/sd1/sweep/44a1dc13c735e657d897/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d59dfba2d230c0666497591b1f9c268adc854fcbba30ca1a5077de08856eaa8
+size 38248
diff --git a/records/sd1/sweep/44a2bde608a3dfa1a828/output.webp b/records/sd1/sweep/44a2bde608a3dfa1a828/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f6cbac582ef76159477dbf739d8debed20809a8e
--- /dev/null
+++ b/records/sd1/sweep/44a2bde608a3dfa1a828/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1968b452eba9f9873d5d8f1815bb8baaf9bf2e84b0768ce87b61d65de6eb0240
+size 18178
diff --git a/records/sd1/sweep/44a3365f07718db15d4b/output.webp b/records/sd1/sweep/44a3365f07718db15d4b/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..aa606352e8f260b1d1c5025daa690e6518cf0702
--- /dev/null
+++ b/records/sd1/sweep/44a3365f07718db15d4b/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77636fa0278e0c698bd8478be2f2899de8e8ec85bc2a24c02707fadede22ca62
+size 22290
diff --git a/records/sd1/sweep/44a80aab4fedee3a51af/output.webp b/records/sd1/sweep/44a80aab4fedee3a51af/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d7b326656fed7d023fa46216000c69be0b2f1b68
--- /dev/null
+++ b/records/sd1/sweep/44a80aab4fedee3a51af/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19914d7852f09698db695e6e45a2c2ea3816fb8cb1d5d3b94c358957e503d8a2
+size 49456
diff --git a/records/sd1/sweep/44aa91b31b6d16187f46/output.webp b/records/sd1/sweep/44aa91b31b6d16187f46/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e3899dc9929740a835be999874cbbc777b402113
--- /dev/null
+++ b/records/sd1/sweep/44aa91b31b6d16187f46/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b0ce5081139e3889d6fcd62642a244f001f24f6486335ad63a96a8265bc7539
+size 44192
diff --git a/records/sd1/sweep/44ad0c24cda69c8c57ba/output.webp b/records/sd1/sweep/44ad0c24cda69c8c57ba/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..8f035f21f5fb4db3c31f6506a384403dddedc7c0
--- /dev/null
+++ b/records/sd1/sweep/44ad0c24cda69c8c57ba/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c89e27161e60f64134f891d503d2ece9b9f8d2e7af56289ee03a488f27c36e6d
+size 22260
diff --git a/records/sd1/sweep/44b94b6757e4f4ef7885/output.webp b/records/sd1/sweep/44b94b6757e4f4ef7885/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..722d11f3ca93049dced1439ce6f45d23949f634f
--- /dev/null
+++ b/records/sd1/sweep/44b94b6757e4f4ef7885/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cac9abafc935a89a1d55c82e7ad8d072a1888cc43decd7fc7e6c28521e2e7ada
+size 45966
diff --git a/records/sd1/sweep/44c18d6ab5023e90fa3b/output.webp b/records/sd1/sweep/44c18d6ab5023e90fa3b/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f0bf23190ab926ea94065f7dfb25769205ca8f72
--- /dev/null
+++ b/records/sd1/sweep/44c18d6ab5023e90fa3b/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b33493060d71a842c308c4ccd3bc4f6edff655400a1b92e987ffa42116d9d8d
+size 19934
diff --git a/records/sd1/sweep/44d690d91235941621fd/output.webp b/records/sd1/sweep/44d690d91235941621fd/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..7e28a2cd2ee5fceccaa9a6d1ddbff721c1db5149
--- /dev/null
+++ b/records/sd1/sweep/44d690d91235941621fd/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9af567d9ee67b01b7b66a0ad8e28dfdd35c45b4714ca337708237460a0230769
+size 18062
diff --git a/records/sd1/sweep/44ec9fa220f7559380e0/output.webp b/records/sd1/sweep/44ec9fa220f7559380e0/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..4a36bc209f289dbf22300432ec15379eb12aaaf4
--- /dev/null
+++ b/records/sd1/sweep/44ec9fa220f7559380e0/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b457698d9e1c48ad2a1bfed3d792273376e2b0482c3ca4fc2b13a886fdcbf2b2
+size 36550
diff --git a/records/sd1/sweep/44f069aa663d3c0bfb3e/output.webp b/records/sd1/sweep/44f069aa663d3c0bfb3e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f24511c988411039e9ef06956e47e23fafba3845
--- /dev/null
+++ b/records/sd1/sweep/44f069aa663d3c0bfb3e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe1236b12378a970ca811a5c75abb5262c2c83718cddb90dc89cfbaabf1f3f1e
+size 32178
diff --git a/records/sd1/sweep/44f4a9e90f669b1c96b0/output.webp b/records/sd1/sweep/44f4a9e90f669b1c96b0/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..bb58a5bbf8a3ec005310de56057c91618c389ea8
--- /dev/null
+++ b/records/sd1/sweep/44f4a9e90f669b1c96b0/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1868c05648d2a6175b24939f7e3567f6c05056191d3e50973fe80f9f854a5222
+size 19728
diff --git a/records/sd1/sweep/44f89eb6e11827698f2f/output.webp b/records/sd1/sweep/44f89eb6e11827698f2f/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0e4e934016ea01ca71095081e1fd4c2d0b59329f
--- /dev/null
+++ b/records/sd1/sweep/44f89eb6e11827698f2f/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:021feb5e919177d509f73a0c1b835e2a736f500b99eef49882bc44016ac9cca1
+size 44356
diff --git a/records/sd1/sweep/451f7cdf1aad5187e202/output.webp b/records/sd1/sweep/451f7cdf1aad5187e202/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..446ef22182dae8bc715cb3fc81a9b528599f3bba
--- /dev/null
+++ b/records/sd1/sweep/451f7cdf1aad5187e202/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:116ff298cf5196ef73ec30baf086585a0b56d3b6b4d2df33cf0d347229b918d3
+size 17842
diff --git a/records/sd1/sweep/451fd0cb4563d433f25f/output.webp b/records/sd1/sweep/451fd0cb4563d433f25f/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..56cce5fc50d8e0800c3138cfdf0d76a4dc4b20f4
--- /dev/null
+++ b/records/sd1/sweep/451fd0cb4563d433f25f/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:300c16156b1b1d1d8d01810f9da34f39b355f739f339f09707e19d24f10b875f
+size 35466
diff --git a/records/sd1/sweep/45286b4d3d47c4208ad2/output.webp b/records/sd1/sweep/45286b4d3d47c4208ad2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d0f4f3655ba7c99666142be015d3b34147212f95
--- /dev/null
+++ b/records/sd1/sweep/45286b4d3d47c4208ad2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1190e719fa5d6935516a1f10c0a3933a0eb6e2f9a0e8487d316759279021b92
+size 34426
diff --git a/records/sd1/sweep/452aa50f8bc1a73ad3e2/output.webp b/records/sd1/sweep/452aa50f8bc1a73ad3e2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d58ea6c44215d2029fe3916789cabfde14f036bb
--- /dev/null
+++ b/records/sd1/sweep/452aa50f8bc1a73ad3e2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8de20c69a0a8dc75f38b7a4241cd0628fbc205f589af431d059650d807ac35e6
+size 36498
diff --git a/records/sd1/sweep/453514cdfdfa2c07c29e/output.webp b/records/sd1/sweep/453514cdfdfa2c07c29e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9f539be500bdb971258d0f1687393584c3aca6ff
--- /dev/null
+++ b/records/sd1/sweep/453514cdfdfa2c07c29e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afe95df299032e6569fa0c15f0bca795313cab0bd378397fcafde439add8b5b8
+size 37838
diff --git a/records/sd1/sweep/4537878e43ac8f1c62a4/output.webp b/records/sd1/sweep/4537878e43ac8f1c62a4/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..11f2fa3474496f1fb561a646391f5510bf0df122
--- /dev/null
+++ b/records/sd1/sweep/4537878e43ac8f1c62a4/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06a790894e7633817b3134c939c9ede56c1e1eea8c4e26b9acd4444b5ab4d2c4
+size 35874
diff --git a/records/sd1/sweep/45378ae63a76a3726b25/output.webp b/records/sd1/sweep/45378ae63a76a3726b25/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ab818cdbbacb3ea6f48f0ccb4109814b61db5376
--- /dev/null
+++ b/records/sd1/sweep/45378ae63a76a3726b25/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be886ad3c33c1aa86b175fc2224a7dae7ee0c0f7d474f1976b401ac62fc69ed7
+size 49170
diff --git a/records/sd1/sweep/453ed9f448822e40c585/output.webp b/records/sd1/sweep/453ed9f448822e40c585/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9d1c5599cc4eb3ddc428b1d589cc4c534780e740
--- /dev/null
+++ b/records/sd1/sweep/453ed9f448822e40c585/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de53c3a135ad855eb8779875b162dc6e187633141dfde8cf5cbcced359f7e3ae
+size 19468
diff --git a/records/sd1/sweep/454b99774105e3a40486/output.webp b/records/sd1/sweep/454b99774105e3a40486/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..bf29bb296ed84a55c29a72f29f2968b193e9fc84
--- /dev/null
+++ b/records/sd1/sweep/454b99774105e3a40486/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3f10e0cc4523ee374f276af5229fb4bbe9785bceffa44591afe06bf37a0349e
+size 36694
diff --git a/records/sd1/sweep/4552aeb2384af8a7d1c2/output.webp b/records/sd1/sweep/4552aeb2384af8a7d1c2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d99f302d9a6eccf8341b324be636fe43c39f99cc
--- /dev/null
+++ b/records/sd1/sweep/4552aeb2384af8a7d1c2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:611b04cbf1f8574d2d5b8369ac3ffe8b5855d0445f4ac39e1b8c7872054ee64d
+size 13578
diff --git a/records/sd1/sweep/45581d3cdb5b161a33c4/output.webp b/records/sd1/sweep/45581d3cdb5b161a33c4/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..eb57fee458b467324849d08c0326e9a2ee65ef1b
--- /dev/null
+++ b/records/sd1/sweep/45581d3cdb5b161a33c4/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e4bc5cf92b6bef6790213df29821ef0c782e7225af76ef1ce5cb391bd49f19a
+size 40658
diff --git a/records/sd1/sweep/455bb596206e36a2f3cb/output.webp b/records/sd1/sweep/455bb596206e36a2f3cb/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..da0246466fdd1df42cad2ab2c0df91149f7d916b
--- /dev/null
+++ b/records/sd1/sweep/455bb596206e36a2f3cb/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cde9f08a7655f74ece62718e3560ffff14b6d2ebbc16e6df5b7c16e65a29186
+size 32250
diff --git a/records/sd1/sweep/455c7281730c9911743a/output.webp b/records/sd1/sweep/455c7281730c9911743a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..45da2be0f96831a5381123c73217171de7cd28d8
--- /dev/null
+++ b/records/sd1/sweep/455c7281730c9911743a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b19311dfd81f8837709a5bf1024d7ec22a3d14db7a5de78a78016940f098a562
+size 58116
diff --git a/records/sd1/sweep/455da1e4fdc620419ba2/output.webp b/records/sd1/sweep/455da1e4fdc620419ba2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ad2ffa1477df1139c3fd56546213c4cdf9f3f3f9
--- /dev/null
+++ b/records/sd1/sweep/455da1e4fdc620419ba2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fef5930453a0c8fae67f227102c1d77212f152360af101c7c1a8c02b7542240a
+size 41996
diff --git a/records/sd1/sweep/456044c309a1659ed9e7/output.webp b/records/sd1/sweep/456044c309a1659ed9e7/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..597504370c6957b2cae4806dd52a0d9c6d8619cc
--- /dev/null
+++ b/records/sd1/sweep/456044c309a1659ed9e7/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:690416f0e242f1e9fa8f8cc497af01c9a71c3793a19b047b30ca364e6cc38cea
+size 19936
diff --git a/records/sd1/sweep/4566e522f6bfea9c3813/output.webp b/records/sd1/sweep/4566e522f6bfea9c3813/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e9d2944131edc0feae1ebf00e3753d7fd0e49e7c
--- /dev/null
+++ b/records/sd1/sweep/4566e522f6bfea9c3813/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:caeea9c22cc93665303ead0ca49d74c84d90f24d4f66a9820574b63c928cac65
+size 27060
diff --git a/records/sd1/sweep/456d5e2441dbf685b053/output.webp b/records/sd1/sweep/456d5e2441dbf685b053/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..875c6b2cc2c550f4175920c1576b9e7d57c2c316
--- /dev/null
+++ b/records/sd1/sweep/456d5e2441dbf685b053/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f83b2e53f5d2c5e9baf3d27a940e1711093c9b2e59fe545b0eca39056b5d45c7
+size 18520
diff --git a/records/sd1/sweep/45761c984bf34f9d4889/output.webp b/records/sd1/sweep/45761c984bf34f9d4889/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..3ed098bcb332f51cb16c1c002065ea6fc63333ec
--- /dev/null
+++ b/records/sd1/sweep/45761c984bf34f9d4889/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:778a5397f92d5f6bfa9ac2654ac9fe48e82b8860e5d2d439d0e7529f23476803
+size 49604
diff --git a/records/sd1/sweep/4588afff7e1838f6e946/output.webp b/records/sd1/sweep/4588afff7e1838f6e946/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..317706e3094c4876e5d651f2867dcc86a9f3f039
--- /dev/null
+++ b/records/sd1/sweep/4588afff7e1838f6e946/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d65126dfcdd2fa1d1e28082c0d79f9c4005248722a0db50da402c682955a4dab
+size 31050
diff --git a/records/sd1/sweep/45a1061655652a257658/output.webp b/records/sd1/sweep/45a1061655652a257658/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0e8de9c708367192ec94bfa4e26f47b5045cc2ac
--- /dev/null
+++ b/records/sd1/sweep/45a1061655652a257658/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34499f5103ea6cb19a08b323c56673b800b61edf17d39120f54b757442e16ee8
+size 37580
diff --git a/records/sd1/sweep/45adf15a3a43a430f43e/output.webp b/records/sd1/sweep/45adf15a3a43a430f43e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..4a248d5b7a100f91015dc97df0ac79db352d4496
--- /dev/null
+++ b/records/sd1/sweep/45adf15a3a43a430f43e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bdf0560d886f80fd5b17b9ac14dcc168615fa0e1fff237f877ebdcea092d24a
+size 29812
diff --git a/records/sd1/sweep/45b2ef7c0241d5012704/output.webp b/records/sd1/sweep/45b2ef7c0241d5012704/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..208296a90241b53f88b16602cac7fe852c2ed600
--- /dev/null
+++ b/records/sd1/sweep/45b2ef7c0241d5012704/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ccb80cc3081e8e986e9bc7eb4dbcfb2605d9a112154a4bf84495e1c9e8d0021
+size 40996
diff --git a/records/sd1/sweep/45b6781bf80c7901c9b8/output.webp b/records/sd1/sweep/45b6781bf80c7901c9b8/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..604c25743fc56b4a176fb9fa5cf5be9030a00999
--- /dev/null
+++ b/records/sd1/sweep/45b6781bf80c7901c9b8/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c3537559b35a93a7269dae888af927a1af7ff25d6a4b8016ae2021e14eded61
+size 24390
diff --git a/records/sd1/sweep/45b6c46ad4014295633c/output.webp b/records/sd1/sweep/45b6c46ad4014295633c/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1f897e46b74719866710dcf6f5b72ce7a8eccf05
--- /dev/null
+++ b/records/sd1/sweep/45b6c46ad4014295633c/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0fc720106260fbeb3de72655646c5eb36e7118a36aeebfe7de2d1590165d10f
+size 53892
diff --git a/records/sd1/sweep/45ccd0d17684b600e148/output.webp b/records/sd1/sweep/45ccd0d17684b600e148/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..33cc4f3e301b3d0f7b1b34a58d1d22c508ae4560
--- /dev/null
+++ b/records/sd1/sweep/45ccd0d17684b600e148/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c767659410e7562949a2dfb91f69b1b350906156235c79c712023b586d4d5f75
+size 30910
diff --git a/records/sd1/sweep/45ccd5564088fcf70be3/output.webp b/records/sd1/sweep/45ccd5564088fcf70be3/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0d25b94b3da22376b3f374ac52624337091b26a9
--- /dev/null
+++ b/records/sd1/sweep/45ccd5564088fcf70be3/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13df35bb8222e40010067694db3bd590c1b1810546d66c8ba3a21ee97890e0d2
+size 14636
diff --git a/records/sd1/sweep/45d0516ab6ffefa19fbd/output.webp b/records/sd1/sweep/45d0516ab6ffefa19fbd/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1c444f6f9cda3705fdec65e70310d9ec76b06ab6
--- /dev/null
+++ b/records/sd1/sweep/45d0516ab6ffefa19fbd/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40ece76435050ae061b9eba8e63390dbe92d7031f27c9ed739bc7a6c876dc713
+size 57664
diff --git a/records/sd1/sweep/45df7251e6c8249867c5/output.webp b/records/sd1/sweep/45df7251e6c8249867c5/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..6bab998ae354dbe97dcbab77d7bdd13c5fdfba5b
--- /dev/null
+++ b/records/sd1/sweep/45df7251e6c8249867c5/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e55d164a2230d06b8757289cdc9b79d0eb3de9307f1904b1858f159fbccf6ee
+size 46126
diff --git a/records/sd1/sweep/45e25f6e606d1dffd926/output.webp b/records/sd1/sweep/45e25f6e606d1dffd926/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..86067102b3853755e9f33a1dc4c5b2fb2e489567
--- /dev/null
+++ b/records/sd1/sweep/45e25f6e606d1dffd926/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fd83ac44c3d75cd35aa777dd568f0074983c87639f89d4c3d16b94da2eb8cb9
+size 24576
diff --git a/records/sd1/sweep/45f2c95aed5dd2feda73/output.webp b/records/sd1/sweep/45f2c95aed5dd2feda73/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c55b1cf5da533416084e2fdc8a2deb3d0bb74df3
--- /dev/null
+++ b/records/sd1/sweep/45f2c95aed5dd2feda73/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd7388f42ab478dace86bca9906fbc6d6bc34481906a49728a1cec9fad8ba5f6
+size 19220
diff --git a/records/sd1/sweep/45f2cfcb62bfea34434f/output.webp b/records/sd1/sweep/45f2cfcb62bfea34434f/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e31bbfd92968a22445f24f5dd8d942c9293d3e5b
--- /dev/null
+++ b/records/sd1/sweep/45f2cfcb62bfea34434f/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22aa73020b0bbfabd3c4bacc87d0c074b706f7f891439d57942b0195a936f973
+size 33988
diff --git a/records/sd1/sweep/46018e4339218a251a32/output.webp b/records/sd1/sweep/46018e4339218a251a32/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0b09ca124df33acd8b9f597daf34f01d8e49e965
--- /dev/null
+++ b/records/sd1/sweep/46018e4339218a251a32/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9174167e064e49af1bf6f79710398eca68232582c2af482ea9c0c5a60dadca41
+size 25862
diff --git a/records/sd1/sweep/461b501825ab2e0c33f6/output.webp b/records/sd1/sweep/461b501825ab2e0c33f6/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..74c5d4efaaef2020acdc73bd189c0f27f4de1151
--- /dev/null
+++ b/records/sd1/sweep/461b501825ab2e0c33f6/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84c033bb83280a7aaca33ff874893ddab63e1cd97fa6f224275f79f042a39b55
+size 25304
diff --git a/records/sd1/sweep/4622adad7851b4ae9e69/output.webp b/records/sd1/sweep/4622adad7851b4ae9e69/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..81d73548524e7bfaa709076a9920483b1ff1e6a4
--- /dev/null
+++ b/records/sd1/sweep/4622adad7851b4ae9e69/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c4715f5ff57ad1d160215f6e4639f390e32064c99d6fe682fafb96467c09985
+size 36494
diff --git a/records/sd1/sweep/46279281cc6a06c327fa/output.webp b/records/sd1/sweep/46279281cc6a06c327fa/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..2460703e76ea20e0a0f6614068eba9c0d75c2f6d
--- /dev/null
+++ b/records/sd1/sweep/46279281cc6a06c327fa/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6219ebc9a6dcab17107f7a6c709eaaf2a7d0c66a0a8312b2f2d29e50a8e1fddb
+size 34514
diff --git a/records/sd1/sweep/4629795e62b12c3aa0c7/output.webp b/records/sd1/sweep/4629795e62b12c3aa0c7/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..10112f36fb90b3bf88c1d5c2884930b5e6b49997
--- /dev/null
+++ b/records/sd1/sweep/4629795e62b12c3aa0c7/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96133f1c50911cc53077ba2b17d63e2e2509b604b71e3f882ef7a3c868f3b8b5
+size 38340
diff --git a/records/sd1/sweep/463ffe06a9600fabdaf0/output.webp b/records/sd1/sweep/463ffe06a9600fabdaf0/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5353f7d96a6f63263fd7cdde0b2bccd538f9357f
--- /dev/null
+++ b/records/sd1/sweep/463ffe06a9600fabdaf0/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:faf974ccd9a36362b825b973d4165545145b2725fa70b4dbf2d0f9633c8bd1d8
+size 21320
diff --git a/records/sd1/sweep/46433496fd2c9d434438/output.webp b/records/sd1/sweep/46433496fd2c9d434438/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..6fbaba3fa5b66847715a46f453aba4429c787d06
--- /dev/null
+++ b/records/sd1/sweep/46433496fd2c9d434438/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a26c3c30bc99368aa3538dc93e0c858c57af52f07828d947ef5f4cf3e6f2038a
+size 30368
diff --git a/records/sd1/sweep/46436538e26719c277c2/output.webp b/records/sd1/sweep/46436538e26719c277c2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..23d11451b6760e6c85f643ad139ea275b690fa95
--- /dev/null
+++ b/records/sd1/sweep/46436538e26719c277c2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e3a6452c710ebc53393edc9be0e8d503f9d9e4230d68cb267e4236b86777e89
+size 31704
diff --git a/records/sd1/sweep/4643f1876bdd899104f3/output.webp b/records/sd1/sweep/4643f1876bdd899104f3/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..54f45f2e4e481f8229821c3a2d7822a9cd75aab1
--- /dev/null
+++ b/records/sd1/sweep/4643f1876bdd899104f3/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce00e14438a8c8b8ee2c9af65ca46491888808befe0df36c5e3bee88af6b4f49
+size 57046
diff --git a/records/sd1/sweep/4649555605248c575c0c/output.webp b/records/sd1/sweep/4649555605248c575c0c/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..25f01b5d248ecd855d73f6e3e4a1c8ffd3e50a19
--- /dev/null
+++ b/records/sd1/sweep/4649555605248c575c0c/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0faf8ae8cc054539748413ad7a538309cfb8bbed71244b8009927e06b431c5af
+size 101870
diff --git a/records/sd1/sweep/465a9f9457f67be3c2f9/output.webp b/records/sd1/sweep/465a9f9457f67be3c2f9/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..6083206a6026344934517536791ffc36eb41dd87
--- /dev/null
+++ b/records/sd1/sweep/465a9f9457f67be3c2f9/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb3e7f6fa34a73cbd52f47b980285327bbaf8c3f6e64dfd9d76eecb0367af870
+size 93774
diff --git a/records/sd1/sweep/46695e7c6966b6fd9099/output.webp b/records/sd1/sweep/46695e7c6966b6fd9099/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..754c79dc4e12bc6ecc1f106f07628dde6143b20c
--- /dev/null
+++ b/records/sd1/sweep/46695e7c6966b6fd9099/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc94dfacc4b7b9798b0066331c6f579ff2d47ec657a5884f9352ded6465021b1
+size 32230
diff --git a/records/sd1/sweep/466c3e1dddce0b536815/output.webp b/records/sd1/sweep/466c3e1dddce0b536815/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..2420481251e8c9206c73c37d2c0cf743e37508d9
--- /dev/null
+++ b/records/sd1/sweep/466c3e1dddce0b536815/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a401b0783ee70255277174aa310293f3152b884702dec4edda42cfde3b5d51c4
+size 6706
diff --git a/records/sd1/sweep/466cce6bb6105a8233aa/output.webp b/records/sd1/sweep/466cce6bb6105a8233aa/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..479b3ca716f5f0d573fada9638af9d114201431c
--- /dev/null
+++ b/records/sd1/sweep/466cce6bb6105a8233aa/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:704afae9058c00f8e9e7fcdd45a83fe57ce12ef6bcac980297a4fa715de50c88
+size 46182
diff --git a/records/sd1/sweep/466f9f86b7249585f37d/output.webp b/records/sd1/sweep/466f9f86b7249585f37d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..833997c6d1dc49087cb9021c5b6c7ad3e5efc083
--- /dev/null
+++ b/records/sd1/sweep/466f9f86b7249585f37d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0ff345f58f6a11c8cb4d4b8856858433034cd7893216c11cd726574c7607444
+size 17734
diff --git a/records/sd1/sweep/468a5178c157d3b4e66a/output.webp b/records/sd1/sweep/468a5178c157d3b4e66a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..117b2a3989dc806e2f82f8134bef01eb966e4a99
--- /dev/null
+++ b/records/sd1/sweep/468a5178c157d3b4e66a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e0eee287a837ffb4ed2edfea15771dadc36afbc30d77e35e7f761b7f5065cf6
+size 34938
diff --git a/records/sd1/sweep/46904af5e964ac038492/output.webp b/records/sd1/sweep/46904af5e964ac038492/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..63d506e602bb97416fe269f0d156d586876756d0
--- /dev/null
+++ b/records/sd1/sweep/46904af5e964ac038492/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f89f847a1c248e4edbf303ab394966ac925b711856eabf23fe0ed44f0edb0de8
+size 25570
diff --git a/records/sd1/sweep/4690b75709313d50b795/output.webp b/records/sd1/sweep/4690b75709313d50b795/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5d1b43fdd719c81f6128edb92e2c75ddc1a26438
--- /dev/null
+++ b/records/sd1/sweep/4690b75709313d50b795/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7bed15c3564e757ba0c59de761ea830d4b15c3d31e5ee10ced23fa5c0842f10
+size 24952
diff --git a/records/sd1/sweep/469adf9a1545c086889b/output.webp b/records/sd1/sweep/469adf9a1545c086889b/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..7830fe78e78c3d009cd3fb61ac059f6ee7ef8fcc
--- /dev/null
+++ b/records/sd1/sweep/469adf9a1545c086889b/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af4049b99622ee1cdfd859ab9f913d0f06cfe8e0fda5337a876542ba71765d74
+size 37382
diff --git a/records/sd1/sweep/469ce5e33be4547e7ba4/output.webp b/records/sd1/sweep/469ce5e33be4547e7ba4/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5a3cf7ef53cc8881e46d027fbc8b3d3178f5ca8c
--- /dev/null
+++ b/records/sd1/sweep/469ce5e33be4547e7ba4/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cd19687e296542768fb0cde1e36d48d5d54614cf24a35bd2118d69604d22279
+size 27586
diff --git a/records/sd1/sweep/46a4332876386e43d377/output.webp b/records/sd1/sweep/46a4332876386e43d377/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..bbfc4b858be3fb635fe8e9e8e091d392c25924fa
--- /dev/null
+++ b/records/sd1/sweep/46a4332876386e43d377/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83a9a5e9d69c56847f681d3910045ea042b93eca36fc84b2e8e824ea2c62838b
+size 95040
diff --git a/records/sd1/sweep/46b878a3f18a949a0cfc/output.webp b/records/sd1/sweep/46b878a3f18a949a0cfc/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c120ae29114239d4a979643c8a35383c446189ea
--- /dev/null
+++ b/records/sd1/sweep/46b878a3f18a949a0cfc/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26e38c3fc8697fa6cce5272435fed16de493182a82b982f97d904e2aa04e6616
+size 81534
diff --git a/records/sd1/sweep/46c5c9100545f80b4fff/output.webp b/records/sd1/sweep/46c5c9100545f80b4fff/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d18681ea002193b1e833db0238daa4a0bf7c8160
--- /dev/null
+++ b/records/sd1/sweep/46c5c9100545f80b4fff/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:863cba34f0a6de1af3df3c60ba6d4f453f57be701876cddf2bc8e0f12cf0a304
+size 33830
diff --git a/records/sd1/sweep/46c61d3960b50d14336e/output.webp b/records/sd1/sweep/46c61d3960b50d14336e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..55d7da3c152fa0a01ec490019a2d13c1c12ac5cf
--- /dev/null
+++ b/records/sd1/sweep/46c61d3960b50d14336e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4b0fb7b14ef08a9981ea334adc8f21ea0ec03c8c93ef7ff3dc37c8bdf743ea2
+size 31370
diff --git a/records/sd1/sweep/46e0e1529059c2b14b2a/output.webp b/records/sd1/sweep/46e0e1529059c2b14b2a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f3c33706f8420c5b3db223e59e61085ec1bb6e1a
--- /dev/null
+++ b/records/sd1/sweep/46e0e1529059c2b14b2a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da0e58698b2c16215855eb7fcf0359a215a41632a69bb9c11dfd3469014f97b8
+size 36260
diff --git a/records/sd1/sweep/46e74c3ec41f0ab7b775/output.webp b/records/sd1/sweep/46e74c3ec41f0ab7b775/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9151eaf23e4c7a28902f7b2bd75e0ebe2939d45e
--- /dev/null
+++ b/records/sd1/sweep/46e74c3ec41f0ab7b775/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e22c49b571712eaaa578860090c24dec4a2affd02c864ae82625437cb463b07
+size 28448
diff --git a/records/sd1/sweep/46ebdace660e93b4f62f/output.webp b/records/sd1/sweep/46ebdace660e93b4f62f/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a2106062e8ed6774ee9743f2f92eb59a213e2bfb
--- /dev/null
+++ b/records/sd1/sweep/46ebdace660e93b4f62f/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a59f086c17de705d0f9f59ca85136c5c8748c7a1bf0e5dd2c69f8e042812dcf
+size 22554
diff --git a/records/sd1/sweep/46edab66e04755ea30a9/output.webp b/records/sd1/sweep/46edab66e04755ea30a9/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a49352210772569401af0e76695acf2ec781ffdb
--- /dev/null
+++ b/records/sd1/sweep/46edab66e04755ea30a9/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61f61c6e23a5b0ba0006ed865aed2070d6f83b61344d20c34e10ce8029ce20f7
+size 23692
diff --git a/records/sd1/sweep/46f115c6702c666d2f5c/output.webp b/records/sd1/sweep/46f115c6702c666d2f5c/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d3d0b37f67276d655c7af83532030c1770466b0e
--- /dev/null
+++ b/records/sd1/sweep/46f115c6702c666d2f5c/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e5713856a1be76d05d04edcf854cb2f10acea79857ff4d290647660189a54b1
+size 37578
diff --git a/records/sd1/sweep/46f2ae59694a03c38cf1/output.webp b/records/sd1/sweep/46f2ae59694a03c38cf1/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e74b3ee6d88f08962cde59283ec09c7f16d84aa8
--- /dev/null
+++ b/records/sd1/sweep/46f2ae59694a03c38cf1/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0322f4c2340a49cd1b8b0b32861700b6822927b99bb18804bcbfbd262d2b99bd
+size 35984
diff --git a/records/sd1/sweep/46fe134d5f0ee7b52de6/output.webp b/records/sd1/sweep/46fe134d5f0ee7b52de6/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..953daeaa70ac154ccf4c963c4ea8558e1382560f
--- /dev/null
+++ b/records/sd1/sweep/46fe134d5f0ee7b52de6/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:217422e36c25be0613d86ab6741bb89c9167f5e4eb1bb04208594a10620a8268
+size 36992
diff --git a/records/sd1/sweep/47017ec81d738186354f/output.webp b/records/sd1/sweep/47017ec81d738186354f/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..b4741469ad91d088acd95bf53d2dc4839b0f25b5
--- /dev/null
+++ b/records/sd1/sweep/47017ec81d738186354f/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc25bcb190950e686fd75cc78898b296b54c9cc2c31c357b98074c89ffba33e5
+size 27154
diff --git a/records/sd1/sweep/470497509a9c397632d6/output.webp b/records/sd1/sweep/470497509a9c397632d6/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f060fca2aaac9207d58c8c57c2d6ee456312741b
--- /dev/null
+++ b/records/sd1/sweep/470497509a9c397632d6/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5a14df9f5ef7c359446022a09062bb86f2f60bd7b702c2e1d8b61088a572881
+size 48776
diff --git a/records/sd1/sweep/470ddf5619b6f964f19c/output.webp b/records/sd1/sweep/470ddf5619b6f964f19c/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..391b2e8a3e0bcd1e78b54ea1984d2db97843b201
--- /dev/null
+++ b/records/sd1/sweep/470ddf5619b6f964f19c/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51a8fc03d7c423009b76507424690ec6a01ece153184922b4b95a97f94b0b2d0
+size 31858
diff --git a/records/sd1/sweep/47174b6158eae7f09e42/output.webp b/records/sd1/sweep/47174b6158eae7f09e42/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..7bd4edf2b3d9b016ea0486fddb9491d8f42b83ac
--- /dev/null
+++ b/records/sd1/sweep/47174b6158eae7f09e42/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7807dbe6480c556aa274d7d3c7087ac4c4d87edbf63a65676da992707dad257d
+size 45140
diff --git a/records/sd1/sweep/471a9b137369de915bbc/output.webp b/records/sd1/sweep/471a9b137369de915bbc/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..3c283294295a4232cb14fc82655b5d61cfe57a70
--- /dev/null
+++ b/records/sd1/sweep/471a9b137369de915bbc/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6630184ee48b49cfd90426d253e21589ca608a3a15fb415eb04162ebbb6ebd72
+size 34020
diff --git a/records/sd1/sweep/471f81c02dd9dc81f9c1/output.webp b/records/sd1/sweep/471f81c02dd9dc81f9c1/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1b6d5cc87dacaba246cb8e1a57cdcf2cb6574c56
--- /dev/null
+++ b/records/sd1/sweep/471f81c02dd9dc81f9c1/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9ff6b1ea38382d1d51817a218dd53c1ec5697c825bbc97cf0a0c4a72e5dd73b
+size 19682
diff --git a/records/sd1/sweep/4720fb5e22057c36209d/output.webp b/records/sd1/sweep/4720fb5e22057c36209d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..755186ee0810f9731bc350aaf1525354d87b21a6
--- /dev/null
+++ b/records/sd1/sweep/4720fb5e22057c36209d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd685a827b2641f126a8e445d24920df79f6bbf4841097801c9c881e9804a150
+size 42674
diff --git a/records/sd1/sweep/472141920bcc4bf4c470/output.webp b/records/sd1/sweep/472141920bcc4bf4c470/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..534e22b6fbf47af47dfef6539eb9ad234fc27e61
--- /dev/null
+++ b/records/sd1/sweep/472141920bcc4bf4c470/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a938916f93c78c9a64a1ad2d3bfc8e9e957b333ce59382174227c6b5106725d3
+size 26556
diff --git a/records/sd1/sweep/473e8f0af3c408fd9a8e/output.webp b/records/sd1/sweep/473e8f0af3c408fd9a8e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..049e290e91f54ea447e0f17c5e01dfabd3e344b7
--- /dev/null
+++ b/records/sd1/sweep/473e8f0af3c408fd9a8e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cec61804942b154d26a1abd3a829d1c597c9cc8d3cdcb716fe9fff1a41e67a3
+size 46150
diff --git a/records/sd1/sweep/4742aec3cc96d7774d9d/output.webp b/records/sd1/sweep/4742aec3cc96d7774d9d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..fe17ecb6c7c548e5d5cbaeac25092331e4b8d689
--- /dev/null
+++ b/records/sd1/sweep/4742aec3cc96d7774d9d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:646c5de1153697d50f45dcdb3a1c28aa6194e3a1ae364d6e26cdba5dbb21f47e
+size 46802
diff --git a/records/sd1/sweep/474dd163154cfeb2e71f/output.webp b/records/sd1/sweep/474dd163154cfeb2e71f/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9f2b53e701d1f196c6d852050248c3d529301fe1
--- /dev/null
+++ b/records/sd1/sweep/474dd163154cfeb2e71f/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b3ae4d4fc54d9adf119d040085649c1be95bf8b9d3673fee9edee86e97baecc
+size 16358
diff --git a/records/sd1/sweep/474ebc8375d01a4a0907/output.webp b/records/sd1/sweep/474ebc8375d01a4a0907/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f273b889f1d0617a27ac7ab585f7a347e5b2fc03
--- /dev/null
+++ b/records/sd1/sweep/474ebc8375d01a4a0907/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80383bbedd0ee9d4d2556a85cfb0caadf2876df957662784324606973dddfd87
+size 28566
diff --git a/records/sd1/sweep/474f3c272c38c942c2b7/output.webp b/records/sd1/sweep/474f3c272c38c942c2b7/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e4d2d0671e904df3c871b8dd0ad10740716c54f5
--- /dev/null
+++ b/records/sd1/sweep/474f3c272c38c942c2b7/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9534b3bb1744557aac01cb36378b1a8741d96e38af3cfcd3583141e55c681a3c
+size 56472
diff --git a/records/sd1/sweep/475cc074ed08aadcec13/output.webp b/records/sd1/sweep/475cc074ed08aadcec13/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..800c879e22c30e9982e12ffc0925d5d11b599ffa
--- /dev/null
+++ b/records/sd1/sweep/475cc074ed08aadcec13/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2aea9acfdac9815c75b7e345e8c846d527606c346e0dfff1a1cb8799578732d2
+size 38104
diff --git a/records/sd1/sweep/4761df11dffefdc3086a/output.webp b/records/sd1/sweep/4761df11dffefdc3086a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a07c1c5b33bfdb425ad9a792075344f7d1e995a7
--- /dev/null
+++ b/records/sd1/sweep/4761df11dffefdc3086a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8abf9db823c9bb4fe84dcedeacb99ecfb67bf6aa510907dc061f223a24197fdf
+size 32882
diff --git a/records/sd1/sweep/47639091e53c475284a9/output.webp b/records/sd1/sweep/47639091e53c475284a9/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..508dd7e1210409654d95b8c9a4c683cb4212047b
--- /dev/null
+++ b/records/sd1/sweep/47639091e53c475284a9/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30d5c890edb5a13c79569f472ee8ac8f1073e501a0e5b9e137931ace87e76c07
+size 42326
diff --git a/records/sd1/sweep/477045ab04e0421a9666/output.webp b/records/sd1/sweep/477045ab04e0421a9666/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..126ac4250d21daae2292946b468f03c3b9e5bb44
--- /dev/null
+++ b/records/sd1/sweep/477045ab04e0421a9666/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ad5a9d712137585c3c6a92f40c87935ffc7e2044e823038b6c095c24070084c
+size 35980
diff --git a/records/sd1/sweep/477ca4c7f51c7cf73501/output.webp b/records/sd1/sweep/477ca4c7f51c7cf73501/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..934be377c5a1eb5fb246f90653dec64992d20899
--- /dev/null
+++ b/records/sd1/sweep/477ca4c7f51c7cf73501/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78cfd5393286760f45149966a7a5eaa8a48395b6543256950a9aee167734fdf0
+size 30310
diff --git a/records/sd1/sweep/47837130ed9fc7c9ba35/output.webp b/records/sd1/sweep/47837130ed9fc7c9ba35/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e1b2fc9de255d37190fc6804029def23da3df76d
--- /dev/null
+++ b/records/sd1/sweep/47837130ed9fc7c9ba35/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8fa54658893cd8985cc19281a8c4773d123af77dd5bed0fff5cc064b281af7b
+size 51362
diff --git a/records/sd1/sweep/479bbf04d34b4100462e/output.webp b/records/sd1/sweep/479bbf04d34b4100462e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9572db559e8671272ca5d56d7953689d12e538a5
--- /dev/null
+++ b/records/sd1/sweep/479bbf04d34b4100462e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52434e7676460d9163e13100d36f32b606e38ac8abf62f306726e29b2d9b99f0
+size 45664
diff --git a/records/sd1/sweep/479c7ec0554c092e41c3/output.webp b/records/sd1/sweep/479c7ec0554c092e41c3/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..b79ac838ed3d426c4efbc0bf420166a87eb6ae76
--- /dev/null
+++ b/records/sd1/sweep/479c7ec0554c092e41c3/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:424d26ddb3a03b4d5b382312d8cdf93be21b44343cfb17eea4f60b415b466449
+size 21670
diff --git a/records/sd1/sweep/479d3809d8786bdb2db3/output.webp b/records/sd1/sweep/479d3809d8786bdb2db3/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..de5c7edf4f031709120a31536e0a44a297c68597
--- /dev/null
+++ b/records/sd1/sweep/479d3809d8786bdb2db3/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3f8a6461e06753d654ea259ea82416d61cf0178e9c908cbecf78a777b87df9f
+size 42702
diff --git a/records/sd1/sweep/479d9dcb6cfa486db58d/output.webp b/records/sd1/sweep/479d9dcb6cfa486db58d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0b4a344006c7ab9c029461c48c74b5ab5b0fe0c1
--- /dev/null
+++ b/records/sd1/sweep/479d9dcb6cfa486db58d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:102cd6aa259d4ea853d79a56690130b0bb4e5761e6c9655f63c968ff99612910
+size 46588
diff --git a/records/sd1/sweep/47a200213b8c0bc87a5e/output.webp b/records/sd1/sweep/47a200213b8c0bc87a5e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d61095c77e23c9cc48668953128bf5c6fb9c9ec1
--- /dev/null
+++ b/records/sd1/sweep/47a200213b8c0bc87a5e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eec817ed3af45369c8942fbd7ca2d35c984a504b9a97173c385aa7e498843803
+size 24488
diff --git a/records/sd1/sweep/47a6f553931b8c3d7a78/output.webp b/records/sd1/sweep/47a6f553931b8c3d7a78/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f95f5685a29a9d086eaae1a3ddb5f213036b98ed
--- /dev/null
+++ b/records/sd1/sweep/47a6f553931b8c3d7a78/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4009fd6168f5af4e545f3675a5b25310acc3a10a646ad790f940e604ad5b6b4
+size 20382
diff --git a/records/sd1/sweep/47a9e8854fdbf8526088/output.webp b/records/sd1/sweep/47a9e8854fdbf8526088/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..71a277613ae9e42f7a6adde9380797e798745e47
--- /dev/null
+++ b/records/sd1/sweep/47a9e8854fdbf8526088/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c2293e07083879dc3317a4f266dde48d76dc14f0eca694ef65d65ead3ae61fb
+size 32644
diff --git a/records/sd1/sweep/47c06d9b322a5e01b345/output.webp b/records/sd1/sweep/47c06d9b322a5e01b345/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1fd708f6c4f232f5a342906df75975ed0a769de1
--- /dev/null
+++ b/records/sd1/sweep/47c06d9b322a5e01b345/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e844941012e63e1357248e2de3e43ac166ac2d5615389fa05adf3917916acd4
+size 37312
diff --git a/records/sd1/sweep/47c8ba16e1069bff58c0/output.webp b/records/sd1/sweep/47c8ba16e1069bff58c0/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..cbd5e34fb4236ebf9592d3b609bd1f89bb0a5cb1
--- /dev/null
+++ b/records/sd1/sweep/47c8ba16e1069bff58c0/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:654bb0efd923a4af0cba9a4dd8e627bbb42d12dae9496452dd9695ceb0e5632c
+size 35512
diff --git a/records/sd1/sweep/47d8c1775d2054a000d1/output.webp b/records/sd1/sweep/47d8c1775d2054a000d1/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ce577471b9767a4873ce3549fb9cc755dc70440c
--- /dev/null
+++ b/records/sd1/sweep/47d8c1775d2054a000d1/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c479f11ff2df654593bac531eedf4267a7b6729c8bfc857304041df61ee8c35e
+size 32324
diff --git a/records/sd1/sweep/47e196e6eefa35aa6f5f/output.webp b/records/sd1/sweep/47e196e6eefa35aa6f5f/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0cdb505c19e7e37c29fa8491d6bd499e63757957
--- /dev/null
+++ b/records/sd1/sweep/47e196e6eefa35aa6f5f/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:949e9cee5b2cdf4f7c7861aa1a71ba0e4ce918c095196d21a824f9183b445471
+size 30676
diff --git a/records/sd1/sweep/47e5a594f1590b27cb8a/output.webp b/records/sd1/sweep/47e5a594f1590b27cb8a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..3a1fc81ec161baeb976e8afe5ff5b1684d54b070
--- /dev/null
+++ b/records/sd1/sweep/47e5a594f1590b27cb8a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f240d4113d0c309b5b5bd9fab13b4b7d03c77391c5db92e408679672e31065b
+size 36466
diff --git a/records/sd1/sweep/47e9bd045790d02791ad/output.webp b/records/sd1/sweep/47e9bd045790d02791ad/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..356d8ebb2bdacd6d672fc6287523747170075c9d
--- /dev/null
+++ b/records/sd1/sweep/47e9bd045790d02791ad/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3c125f4af7f4ad5e9075add39456067226d4a4f3eadf0a6177c4a0e9596028c
+size 50484
diff --git a/records/sd1/sweep/47f7a492e89b93a9e614/output.webp b/records/sd1/sweep/47f7a492e89b93a9e614/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..7b60af4552de213d9195cfd9ba3fa528643faab8
--- /dev/null
+++ b/records/sd1/sweep/47f7a492e89b93a9e614/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8fb9a7b87313a4ef29eb15b9c59ac87a3df8e697c705d3a0d595ef01d87bda97
+size 20994
diff --git a/records/sd1/sweep/47f8101835bec67ec7d1/output.webp b/records/sd1/sweep/47f8101835bec67ec7d1/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..4b048c283b498f4006203c50dba83cf867ae8eaf
--- /dev/null
+++ b/records/sd1/sweep/47f8101835bec67ec7d1/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60ff6207dff171771c4034e4574fe88783a30a1bc5eed6976e9e8d37f47fae06
+size 23952
diff --git a/records/sd1/sweep/47f818cd9c0f2c2cd049/output.webp b/records/sd1/sweep/47f818cd9c0f2c2cd049/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..b2cf8fadb2ae33261ad44a80fc0e0dff7e59ecff
--- /dev/null
+++ b/records/sd1/sweep/47f818cd9c0f2c2cd049/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4aaa3e1c96609dff45340da5039ee334335f82b9565f45b97d5de1a78f36634
+size 20770
diff --git a/records/sd1/sweep/480305b3b981925dbf8a/output.webp b/records/sd1/sweep/480305b3b981925dbf8a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..bde6dc35f4382b9aac6b9ac810fbe2809fe197f4
--- /dev/null
+++ b/records/sd1/sweep/480305b3b981925dbf8a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:265ca3457673b221c5f72f7499b9323551e697a12ea92ee569169ce764f858c5
+size 38696
diff --git a/records/sd1/sweep/480461fe454bf7214b8b/output.webp b/records/sd1/sweep/480461fe454bf7214b8b/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..90be2bfb4e471928b421c4cfb728ed209439410e
--- /dev/null
+++ b/records/sd1/sweep/480461fe454bf7214b8b/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46f57c08fefb6abd30243fb81fb63227df1eb01debe7dd3cb7ea9a63d464d5bf
+size 37608
diff --git a/records/sd1/sweep/480b963a1898c327b4ef/output.webp b/records/sd1/sweep/480b963a1898c327b4ef/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5ecf6dca987386997e27a36269dc6d3de21f93af
--- /dev/null
+++ b/records/sd1/sweep/480b963a1898c327b4ef/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de1093c1dcdc378bda08e45aa3ba9ad1e93c2a2ecf30e0742d35b5c655cc47fe
+size 48390
diff --git a/records/sd1/sweep/480c2666bb99c0141f02/output.webp b/records/sd1/sweep/480c2666bb99c0141f02/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e5f989eeac43673e3bedca6830a4f5fe67f8c340
--- /dev/null
+++ b/records/sd1/sweep/480c2666bb99c0141f02/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eac236803288ee4ddb1da8d822c21dd59f9993b363afebecf40623ff10c48100
+size 31418
diff --git a/records/sd1/sweep/482416f2fc308a6e321d/output.webp b/records/sd1/sweep/482416f2fc308a6e321d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a2446a7d0a1d5daa29890544663a39054d583f4e
--- /dev/null
+++ b/records/sd1/sweep/482416f2fc308a6e321d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b64ecfcc4eb8b8dab4dc9fb49a33090f7de3ce540b8d1e82d5a3bb20e5475f1
+size 31738
diff --git a/records/sd1/sweep/4833360ac7347effb7f3/output.webp b/records/sd1/sweep/4833360ac7347effb7f3/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..6a561537cd8442aca2a45e85376f9a42dfd61e4a
--- /dev/null
+++ b/records/sd1/sweep/4833360ac7347effb7f3/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efd9a2bda0b52544080d96d0e27083c3823a060a09cd164191d1e2568a2df8eb
+size 21788
diff --git a/records/sd1/sweep/48341324367f2bbfee33/output.webp b/records/sd1/sweep/48341324367f2bbfee33/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9d050e72dd94708515b55cf20dfd6369ff45fba7
--- /dev/null
+++ b/records/sd1/sweep/48341324367f2bbfee33/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b72164c8bfe875304a0ca137cd553b11998fc15526b33773b0be7d9783ffcbe2
+size 22088
diff --git a/records/sd1/sweep/48392157240c7b550a32/output.webp b/records/sd1/sweep/48392157240c7b550a32/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..27a4833785ebcc123436d74095b99bb55bcf522f
--- /dev/null
+++ b/records/sd1/sweep/48392157240c7b550a32/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f2fef4692b5989e081b1cedcfb69e02f3915af83da2d660ef1fefa7f84d3825
+size 40538
diff --git a/records/sd1/sweep/484248ad7b944687fa77/output.webp b/records/sd1/sweep/484248ad7b944687fa77/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a2f3b77d0e2af6b4971dacaa6e90d02db4df4a64
--- /dev/null
+++ b/records/sd1/sweep/484248ad7b944687fa77/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5dee99b3a3cef1e0c3104665e7f2a9b385d3fabacf5e5b5b66eb956aa9589c0
+size 32326
diff --git a/records/sd1/sweep/484eed7eeee6b2fdc9e5/output.webp b/records/sd1/sweep/484eed7eeee6b2fdc9e5/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ba52decc55c30ceea713d4e9f3e35af7e03be940
--- /dev/null
+++ b/records/sd1/sweep/484eed7eeee6b2fdc9e5/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36cb071654e54b4a7572c4ef954a214b622691b01600cd160ad1e772e36ae269
+size 23974
diff --git a/records/sd1/sweep/4858eea5fd3ab9d5092c/output.webp b/records/sd1/sweep/4858eea5fd3ab9d5092c/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..2a2d1d5ff62b9b8d41b06a87061e0d4707a52191
--- /dev/null
+++ b/records/sd1/sweep/4858eea5fd3ab9d5092c/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30fb82b45aed85319b2bd83c288d18af9496da7abc8fc3956b2d43540203aaf0
+size 26604
diff --git a/records/sd1/sweep/48608cecb0e1cb52e461/output.webp b/records/sd1/sweep/48608cecb0e1cb52e461/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..84bab0fb2257ba023adc984231f2bb3ebc12465d
--- /dev/null
+++ b/records/sd1/sweep/48608cecb0e1cb52e461/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26226625c4e9399a9cf9ea305fbdab272e42fd3e1a12839f8a7f136c485a15d2
+size 102672
diff --git a/records/sd1/sweep/48616c06ea27870e76fb/output.webp b/records/sd1/sweep/48616c06ea27870e76fb/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..344db754c7868faacc42336a97b3875a7d257f65
--- /dev/null
+++ b/records/sd1/sweep/48616c06ea27870e76fb/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31095c27abb0e794f3881705813511c76212600aed93c182ebbb1207a86313d2
+size 57534
diff --git a/records/sd1/sweep/486581aa372a182dbba8/output.webp b/records/sd1/sweep/486581aa372a182dbba8/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..b2872e85208ca3423583ed8b691a85c05bee0703
--- /dev/null
+++ b/records/sd1/sweep/486581aa372a182dbba8/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4dd0504111415b6d26841fdfa6012de5a2ab325243f80439fb809c94cbff6222
+size 38234
diff --git a/records/sd1/sweep/486e9b968e476e27f2ec/output.webp b/records/sd1/sweep/486e9b968e476e27f2ec/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..70b38634f1509812b65b3fc356f9d6d550494f59
--- /dev/null
+++ b/records/sd1/sweep/486e9b968e476e27f2ec/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc6b2f7c107332e6dc7c05c8a8c72bba13f040d789ab64d30597beaaacb4e76b
+size 23424
diff --git a/records/sd1/sweep/4872e8ff6b0ede588cc6/output.webp b/records/sd1/sweep/4872e8ff6b0ede588cc6/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a91eba8b7a9c2d104aa02914634e7284a6a564f2
--- /dev/null
+++ b/records/sd1/sweep/4872e8ff6b0ede588cc6/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf5fb1dd4a413ecd257be528e9e7904c66c1b4bbfdfea9bf7b255842b74447ba
+size 25222
diff --git a/records/sd1/sweep/48839def207951b99033/output.webp b/records/sd1/sweep/48839def207951b99033/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..13f1de63bccb1ef5e17453a42e2c97acc6d67c9a
--- /dev/null
+++ b/records/sd1/sweep/48839def207951b99033/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1554f224e6864a0feb6782ab64769d772549cd0e765344c4ef6e308dd8d064b0
+size 46376
diff --git a/records/sd1/sweep/488554bce7305af80f1f/output.webp b/records/sd1/sweep/488554bce7305af80f1f/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5cef5f184574e441b705ac171afaccc5e8efe07b
--- /dev/null
+++ b/records/sd1/sweep/488554bce7305af80f1f/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:789a5fae740853ed1ea6b0b3be962b78003d09d6702dd372021a20ee9b77401d
+size 26022
diff --git a/records/sd1/sweep/489c72ce28155eb2640b/output.webp b/records/sd1/sweep/489c72ce28155eb2640b/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..4cf1b95ac7647ab929f54f69a5a7e20fe7410de0
--- /dev/null
+++ b/records/sd1/sweep/489c72ce28155eb2640b/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b24ca90765131bf3c2391ca703d450cf963494bd9d2c8b76bbb1a0d517aefcec
+size 101340
diff --git a/records/sd1/sweep/48ab2f9d753e36286a7a/output.webp b/records/sd1/sweep/48ab2f9d753e36286a7a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..53967cdd6da118b739bdcd993b44f44fbe58ef10
--- /dev/null
+++ b/records/sd1/sweep/48ab2f9d753e36286a7a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:419d9118d99627aa242b939b378f2fe500991c408760d8d2ed6aa8aec644bf9d
+size 38504
diff --git a/records/sd1/sweep/48ac628c9434de764bc5/output.webp b/records/sd1/sweep/48ac628c9434de764bc5/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c23d7e0ffb7f90d4680dc4b68034898a1dbf0910
--- /dev/null
+++ b/records/sd1/sweep/48ac628c9434de764bc5/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf932e179bddf9ba23c1bc1871080e86831f9c9a55f3e9f10a0ff04b39558de4
+size 36498
diff --git a/records/sd1/sweep/48ad44e566b1999b83c1/output.webp b/records/sd1/sweep/48ad44e566b1999b83c1/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a0dc606d6734b602b03412257c9a3e051ac35356
--- /dev/null
+++ b/records/sd1/sweep/48ad44e566b1999b83c1/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddac1dcdbe9795590aa6ff56c9aa79e913193581e042071fe2bf153add980c15
+size 67348
diff --git a/records/sd1/sweep/48ad5a2c3dff4b368b44/output.webp b/records/sd1/sweep/48ad5a2c3dff4b368b44/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a18bc66fbcad7d638613ae0a2fd1ed605e793367
--- /dev/null
+++ b/records/sd1/sweep/48ad5a2c3dff4b368b44/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b18bc55dad201f30f938033080985924802792ff1071778cf4bf3d3f70eeb04
+size 33736
diff --git a/records/sd1/sweep/48b370728502c4ef06d5/output.webp b/records/sd1/sweep/48b370728502c4ef06d5/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..34243ed126af181b5846e3d0873f05bf6ec7e369
--- /dev/null
+++ b/records/sd1/sweep/48b370728502c4ef06d5/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:123710d6ae9168baf28eb8cae8d6061cc459d4b853af952d64654ba7f5ff3189
+size 18218
diff --git a/records/sd1/sweep/48bcd0cd4b4f95205b00/output.webp b/records/sd1/sweep/48bcd0cd4b4f95205b00/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..5cecdb72498bcde71dbf491260fa455204371129
--- /dev/null
+++ b/records/sd1/sweep/48bcd0cd4b4f95205b00/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e985bcd61d8ebf810355d336e8314348913a9ee11f38dceaa11ae6b1321ac6e
+size 36502
diff --git a/records/sd1/sweep/48c2f9e0345fdc77cf41/output.webp b/records/sd1/sweep/48c2f9e0345fdc77cf41/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..3fac992c2f26bff5b90b9e25094e9cfb48b71bee
--- /dev/null
+++ b/records/sd1/sweep/48c2f9e0345fdc77cf41/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c46ddedfa47bb2b3b27df9fa0383473a57493827a99df1f8221bbde556875b22
+size 37182
diff --git a/records/sd1/sweep/48c51602ff3cca56712a/output.webp b/records/sd1/sweep/48c51602ff3cca56712a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d8229feef66062e96a09001fc6754121922ac892
--- /dev/null
+++ b/records/sd1/sweep/48c51602ff3cca56712a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3d20170022a53f3c2021b880b98aca934f62b71163d9ab04aa2d963e84bc2fd
+size 32618
diff --git a/records/sd1/sweep/48d329ac3c50c1269368/output.webp b/records/sd1/sweep/48d329ac3c50c1269368/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..be98c654cb6c76da30491cf084f838d2e1535d0a
--- /dev/null
+++ b/records/sd1/sweep/48d329ac3c50c1269368/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:417cb2f08e0d45b4e71ac016c5cecffcb541a3943331915ec176d6a2fa0ecebc
+size 23680
diff --git a/records/sd1/sweep/48d6df996f3604c13a65/output.webp b/records/sd1/sweep/48d6df996f3604c13a65/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1cccb8960b01eeed6433e557cebb86a1b9c83e20
--- /dev/null
+++ b/records/sd1/sweep/48d6df996f3604c13a65/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a9c6765ed27922ec48e7eb5427119e9475ddbf5f23ebde04339c0f5eb4d0c39
+size 40174
diff --git a/records/sd1/sweep/48d91086d9642fa4ce6d/output.webp b/records/sd1/sweep/48d91086d9642fa4ce6d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c67be1ba1424a22a5a2ca95c7b173057dbc5940e
--- /dev/null
+++ b/records/sd1/sweep/48d91086d9642fa4ce6d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e62a8cdaa496a6b41cb3ad281ee885b647b8d86a005eff2ec56bc9be2608f1fb
+size 19934
diff --git a/records/sd1/sweep/48ddf86ebc96f70e7d7a/output.webp b/records/sd1/sweep/48ddf86ebc96f70e7d7a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..9ce45225348108546af388758d346ea4b1082848
--- /dev/null
+++ b/records/sd1/sweep/48ddf86ebc96f70e7d7a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7336fedcffe417f5276db98ef3c4fdd2305ba66599bd3c115f63b00669ed1b68
+size 51064
diff --git a/records/sd1/sweep/48e17eb9add3281db08a/output.webp b/records/sd1/sweep/48e17eb9add3281db08a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..eb22f62c2f18f2f9317ac46399daeeebecc4d8ca
--- /dev/null
+++ b/records/sd1/sweep/48e17eb9add3281db08a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:148f9b765faf85554110a5af64180b011c522b13e959a75775e21481bec3a7d5
+size 50200
diff --git a/records/sd1/sweep/48e3d3db40022fe4552b/output.webp b/records/sd1/sweep/48e3d3db40022fe4552b/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1d8826f62a9cf27f8c6ef2100c9b2fe1e16f7649
--- /dev/null
+++ b/records/sd1/sweep/48e3d3db40022fe4552b/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81fd00bd2827f4aac123da4b03bfa49e54d4a9a085e2ac29c35b7b157f668450
+size 30444
diff --git a/records/sd1/sweep/48f6a1965c4efd42baa2/output.webp b/records/sd1/sweep/48f6a1965c4efd42baa2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..03a86f84982f2880bc0ce3550c6f864a1dce15e4
--- /dev/null
+++ b/records/sd1/sweep/48f6a1965c4efd42baa2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb3c0dbe8cb401ff08cc4e0b54934d7c252821f60cd56692c7761c2536283051
+size 44974
diff --git a/records/sd1/sweep/49016264aaa7ff101261/output.webp b/records/sd1/sweep/49016264aaa7ff101261/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..da58e76e57ef3dc82a432c291b157daf1578b464
--- /dev/null
+++ b/records/sd1/sweep/49016264aaa7ff101261/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a602563e5307b50e50aeb1891e2fe1c849d83aec753708c72526441130390f9
+size 24630
diff --git a/records/sd1/sweep/490fe2966eddaf10a04d/output.webp b/records/sd1/sweep/490fe2966eddaf10a04d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d164341e579bd053c4855ad5fb5508032fae5f27
--- /dev/null
+++ b/records/sd1/sweep/490fe2966eddaf10a04d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f52ca1791a71c46bb8e8b1486aa6061d9b1b777b6368e5c3c4a54325f3f1bdca
+size 34148
diff --git a/records/sd1/sweep/49117a6ebbd4b2c743e0/output.webp b/records/sd1/sweep/49117a6ebbd4b2c743e0/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e3d46a7798c24c2d503d7b8a5ad62ddfcb6c3d73
--- /dev/null
+++ b/records/sd1/sweep/49117a6ebbd4b2c743e0/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2df2c0860602029c891f0cd46585cd9220bc823e526a6b98722eca6458ca8d9f
+size 19936
diff --git a/records/sd1/sweep/491840bee6732eec87a7/output.webp b/records/sd1/sweep/491840bee6732eec87a7/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..dae26539509feb1cba1968fd6c5308aa24c27ee8
--- /dev/null
+++ b/records/sd1/sweep/491840bee6732eec87a7/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d8ff4e23ad76df6dad3517ed36183210ad883b16355b635c3d47740a7678ee2
+size 23992
diff --git a/records/sd1/sweep/491ed2d1b8ad5c3440dd/output.webp b/records/sd1/sweep/491ed2d1b8ad5c3440dd/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..de40dccad095ca8e082024d53d4163e7ed0a94b1
--- /dev/null
+++ b/records/sd1/sweep/491ed2d1b8ad5c3440dd/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1653b33a65b12820e7dbf447e470027703caa39546d0dbb817b7bc7959eb71a1
+size 22476
diff --git a/records/sd1/sweep/49224cde356c5c8e68eb/output.webp b/records/sd1/sweep/49224cde356c5c8e68eb/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..8e2ecff4bc58fe44c60fe2b7fd08917b1af28f36
--- /dev/null
+++ b/records/sd1/sweep/49224cde356c5c8e68eb/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e807d6285a000085f22ba2fe34e3b32ff4de70291262a5fd8a4d873a03007fe6
+size 48990
diff --git a/records/sd1/sweep/493e078f1b059c306154/output.webp b/records/sd1/sweep/493e078f1b059c306154/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..39e702c387f204f38080b570115cc9d459557cba
--- /dev/null
+++ b/records/sd1/sweep/493e078f1b059c306154/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:565ac73e135310a7574acfc494ed10e5382f8f35d0c85bd08981b7bcd5dd5a9f
+size 27480
diff --git a/records/sd1/sweep/493ed8d2463517933b78/output.webp b/records/sd1/sweep/493ed8d2463517933b78/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..33af04ec98dc35c209185868b9c05a42ac1fd001
--- /dev/null
+++ b/records/sd1/sweep/493ed8d2463517933b78/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b29ba714ff9d48d1d2cf2b7a5cae6881a282c54fb9ef2b793aa0c300033c141c
+size 15622
diff --git a/records/sd1/sweep/4947787bc480a73fef62/output.webp b/records/sd1/sweep/4947787bc480a73fef62/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..6308b0415642392ab3fd68daa27c3bcb92b3386f
--- /dev/null
+++ b/records/sd1/sweep/4947787bc480a73fef62/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3248d7015097c7a33fc936729b64b6b21200e54d70f41235013211143d3553a
+size 52322
diff --git a/records/sd1/sweep/494afcfcc029ded8714c/output.webp b/records/sd1/sweep/494afcfcc029ded8714c/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1c3453ccfa343ad3bb67dcc16837f6a563d614d5
--- /dev/null
+++ b/records/sd1/sweep/494afcfcc029ded8714c/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6e0e2fa23d69d6ef9afcab338282abdc929ef72d20c2abe04ac2279fadcfce3
+size 59290
diff --git a/records/sd1/sweep/494de500023a8c180e90/output.webp b/records/sd1/sweep/494de500023a8c180e90/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ac5960938f4de857dc02ef1ca939356eb34b0117
--- /dev/null
+++ b/records/sd1/sweep/494de500023a8c180e90/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:238adc0b90815eaf0b448bcca16948350bed4835a0f6ee7eb04b193eb7c218c8
+size 37984
diff --git a/records/sd1/sweep/495d1863c2696351a4c0/output.webp b/records/sd1/sweep/495d1863c2696351a4c0/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..45bbde5b2469faedcc3ee3b4521f09f0e111331b
--- /dev/null
+++ b/records/sd1/sweep/495d1863c2696351a4c0/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99c7dd783d5fb81ca36d56a2f62d9c313ee22dbeaa7be8cd3ee682acf7858651
+size 44672
diff --git a/records/sd1/sweep/4962daa721f8ff0ac451/output.webp b/records/sd1/sweep/4962daa721f8ff0ac451/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..884dcbba559fea52f0e82cf3a8ccbbab48eed9c0
--- /dev/null
+++ b/records/sd1/sweep/4962daa721f8ff0ac451/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6dbc879a0c1fe91616d5dec4fb3d070a1012f1d771eb895dd3b414eefba228a5
+size 9916
diff --git a/records/sd1/sweep/496bb3fd4234c7096b18/output.webp b/records/sd1/sweep/496bb3fd4234c7096b18/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d138103200d9a44b1ec35a6d65ef2488897b7889
--- /dev/null
+++ b/records/sd1/sweep/496bb3fd4234c7096b18/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9140df39b893c578c79b465d2f47e6cdaebd3c31693ad0e84f7381bdeb79a1f4
+size 25448
diff --git a/records/sd1/sweep/49736deafa87282ceedb/output.webp b/records/sd1/sweep/49736deafa87282ceedb/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e37536065b18781908c072e3ffca3b410cf83a6a
--- /dev/null
+++ b/records/sd1/sweep/49736deafa87282ceedb/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:121e356fa12c026d7dc53bc0764b0e969b636edb29e811c134997d3b63ae6c59
+size 46348
diff --git a/records/sd1/sweep/4978cca3cc9300ea829d/output.webp b/records/sd1/sweep/4978cca3cc9300ea829d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..89ffbeeb5d55a82693daf5397184baa1d5a241d7
--- /dev/null
+++ b/records/sd1/sweep/4978cca3cc9300ea829d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db8343a56e0087eda0e639f421c8a8e9edeaa38a7f57ee8626025e6722747396
+size 55032
diff --git a/records/sd1/sweep/4979e460f81d928b48b8/output.webp b/records/sd1/sweep/4979e460f81d928b48b8/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1597690b23b116d488439a468e79403701ddeb72
--- /dev/null
+++ b/records/sd1/sweep/4979e460f81d928b48b8/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8fce8be2c3e735769ccf4256f197ab1b9d092763a4f6bb667f0a698b7568e89f
+size 62860
diff --git a/records/sd1/sweep/497f7c262f6b8df7a8e0/output.webp b/records/sd1/sweep/497f7c262f6b8df7a8e0/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f2db0d42c139f3df9843abf329f8e21672cb86d0
--- /dev/null
+++ b/records/sd1/sweep/497f7c262f6b8df7a8e0/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9550adf0dc05e6e9ce2678a829eade725289c7e4d6bfddec16f79487d52eb959
+size 15862
diff --git a/records/sd1/sweep/4984143b5496e1de359e/output.webp b/records/sd1/sweep/4984143b5496e1de359e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..7cd03038f1c4a3c146c5a7349bd905cbceac0aef
--- /dev/null
+++ b/records/sd1/sweep/4984143b5496e1de359e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b816f61f04cfbde44c75927ef983ba119f92f5c7b7ca4ddb71fb0b1272e43b1
+size 15666
diff --git a/records/sd1/sweep/49846846ed1ad00a2a23/output.webp b/records/sd1/sweep/49846846ed1ad00a2a23/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..8e12e7cdd376f27d69b6a96b484435d929fafde4
--- /dev/null
+++ b/records/sd1/sweep/49846846ed1ad00a2a23/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0aa28bb619814d98de903da38f40d4e8ce325e72c71fb1c6faa83da7064e7bb
+size 31360
diff --git a/records/sd1/sweep/498e961b1b625213e9b6/output.webp b/records/sd1/sweep/498e961b1b625213e9b6/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..f3a12309fe6c8e1c2bb7d41fa02bffc1cb1a9a98
--- /dev/null
+++ b/records/sd1/sweep/498e961b1b625213e9b6/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b759cfcfca7eb13f8c060fe37538a6dbca47c6726d8e93c7f4cbe299f3cebf9
+size 26972
diff --git a/records/sd1/sweep/4992b83e0cb4100334ca/output.webp b/records/sd1/sweep/4992b83e0cb4100334ca/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..fffebc96884e7acd7defe74db2bfff7e5bfe8934
--- /dev/null
+++ b/records/sd1/sweep/4992b83e0cb4100334ca/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b006d2bcefbff7308b58b152c4e8d34ba21ccbe49914cc30688a6eff8e7e5c4
+size 38120
diff --git a/records/sd1/sweep/499ba0d48f0e3257896d/output.webp b/records/sd1/sweep/499ba0d48f0e3257896d/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..6b2b167450b96bbd79d166f057b7e4372bcd4637
--- /dev/null
+++ b/records/sd1/sweep/499ba0d48f0e3257896d/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfa85e2d2e7f536646531b6c47c7c61d41db6ab53a79a8c88ce067acdd67ae39
+size 45768
diff --git a/records/sd1/sweep/49a08d2e65986d7729b4/output.webp b/records/sd1/sweep/49a08d2e65986d7729b4/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..2e543edfaf4ab6b0970db78ae6d0af0bf1c3f42d
--- /dev/null
+++ b/records/sd1/sweep/49a08d2e65986d7729b4/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:841a9323431cf7e5f1e9af956863497f76b4831ce0c41fe8b250692d2fdb6498
+size 25534
diff --git a/records/sd1/sweep/49b2d19df90bac3814c5/output.webp b/records/sd1/sweep/49b2d19df90bac3814c5/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..98042a7ecf39cd98b79db8c54c33d89cabb9860c
--- /dev/null
+++ b/records/sd1/sweep/49b2d19df90bac3814c5/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9a5d7fbf80083aeb80cfa2f0c43f91fd3d4903f7c831191edc1fd41bbd0379e
+size 36504
diff --git a/records/sd1/sweep/49b6d87a03e53b95ffb2/output.webp b/records/sd1/sweep/49b6d87a03e53b95ffb2/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0cdb4e79a4f46f08a9713116b46ddf8ca90ef946
--- /dev/null
+++ b/records/sd1/sweep/49b6d87a03e53b95ffb2/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:623fc6c442a6416e8ea762c23071236c4632502bbf86a4308047d328d3c318b6
+size 37334
diff --git a/records/sd1/sweep/49b800967702eb971fcf/output.webp b/records/sd1/sweep/49b800967702eb971fcf/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..10d103080530eac27002d5085ec9e6212804e510
--- /dev/null
+++ b/records/sd1/sweep/49b800967702eb971fcf/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b115028038cc859b3d8fc1b4cab510f6f337d4d5de8f071608fd4cd9197e805c
+size 24856
diff --git a/records/sd1/sweep/49b8f3244192b3b8cdf4/output.webp b/records/sd1/sweep/49b8f3244192b3b8cdf4/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c187c4f24997d8ac3223ff218fe8f3c323dadba6
--- /dev/null
+++ b/records/sd1/sweep/49b8f3244192b3b8cdf4/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:552fd049801a879828a4a2541d3979b949913ee26b47430146801335429bd2e1
+size 19652
diff --git a/records/sd1/sweep/49c819e0f4a64d50fe17/output.webp b/records/sd1/sweep/49c819e0f4a64d50fe17/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..d79890c77b1b1f37d6ad94b7f798d75844c40e6e
--- /dev/null
+++ b/records/sd1/sweep/49c819e0f4a64d50fe17/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e7b3ff93e005990fc32e7e56feeb8f19564cda7a1c5ec73ec34c8ae6850d02c
+size 42960
diff --git a/records/sd1/sweep/49cd877bbaa7db133edf/output.webp b/records/sd1/sweep/49cd877bbaa7db133edf/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..50d81a6e783e49887ca891653ee0e2294fb1ae89
--- /dev/null
+++ b/records/sd1/sweep/49cd877bbaa7db133edf/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e21666f3690bf17e7b135c31abe7677a79d33daf77492c141b208efb224f9a8
+size 38420
diff --git a/records/sd1/sweep/49d38331f1d55d8cc3de/output.webp b/records/sd1/sweep/49d38331f1d55d8cc3de/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ad116ed088540398c66f36b6487085150d95f93a
--- /dev/null
+++ b/records/sd1/sweep/49d38331f1d55d8cc3de/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61adb2f6a9b8e0aaf8fb985d255d700682d066c8ea625ccac258d44a8754b950
+size 37072
diff --git a/records/sd1/sweep/49d3a34714c5f45bf979/output.webp b/records/sd1/sweep/49d3a34714c5f45bf979/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..e7343faad0904102aeb54e6d350dcc472d417828
--- /dev/null
+++ b/records/sd1/sweep/49d3a34714c5f45bf979/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:741835076dbdd2d88006f4baf05d4dcf795c121cbe587b259d57f40757e80678
+size 45148
diff --git a/records/sd1/sweep/49e5a3423afcc7e54e7a/output.webp b/records/sd1/sweep/49e5a3423afcc7e54e7a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..881991f6f784f9088a1ecb8aca0077f80264c261
--- /dev/null
+++ b/records/sd1/sweep/49e5a3423afcc7e54e7a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:625d5c0685404c2808e120190065e0e4d326d4fb763b2e5b13e26bd7ae1b74dd
+size 35318
diff --git a/records/sd1/sweep/49e6605533ce6339ff73/output.webp b/records/sd1/sweep/49e6605533ce6339ff73/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..1da168564376cfeacbdb28e0fb0aec2ee97be510
--- /dev/null
+++ b/records/sd1/sweep/49e6605533ce6339ff73/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3584bd221435b6fb68c079d8b3016b593ffb84ea80e2d8b81607b9648becc5f4
+size 27048
diff --git a/records/sd1/sweep/49eea2eee45f8a9463e6/output.webp b/records/sd1/sweep/49eea2eee45f8a9463e6/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..049649b3954c12d7989cff136cabd89fe90c67cb
--- /dev/null
+++ b/records/sd1/sweep/49eea2eee45f8a9463e6/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c1c6697f2b2cf2463aa18effd833e80999ac8c0690db6dcb1d01179cc9250a0
+size 19934
diff --git a/records/sd1/sweep/49fac87e9bbadea86c57/output.webp b/records/sd1/sweep/49fac87e9bbadea86c57/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..7ca42ae813bffd2456e5ab90d3d8d1c4b2725e4a
--- /dev/null
+++ b/records/sd1/sweep/49fac87e9bbadea86c57/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb446ad1a1ed5fb96870a032598d734e1d323ea579aaaec16e2e76ddf85baf1b
+size 15254
diff --git a/records/sd1/sweep/49fc69186b007b02ab89/output.webp b/records/sd1/sweep/49fc69186b007b02ab89/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..2c499f587c1a64340467a3aecd33971f5b7c1fff
--- /dev/null
+++ b/records/sd1/sweep/49fc69186b007b02ab89/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da11cec2d716322b4fc92cebc71ccd102b528503454cc66985e2bb35a485c736
+size 17036
diff --git a/records/sd1/sweep/49ffcaebb23fafc17b9e/output.webp b/records/sd1/sweep/49ffcaebb23fafc17b9e/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..32a2ea1062ae9ac37331c5de50a3a58aa9fd44e0
--- /dev/null
+++ b/records/sd1/sweep/49ffcaebb23fafc17b9e/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b4ce0041cb856d659a5a15ec72df2ae075c2eecf61e7b366a4af3f99735a583
+size 30834
diff --git a/records/sd1/sweep/4a05748437cbb843d88a/output.webp b/records/sd1/sweep/4a05748437cbb843d88a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ed4e8618a8f15529f2d667be0f64a9e2b0c3671c
--- /dev/null
+++ b/records/sd1/sweep/4a05748437cbb843d88a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a33fcb64a7a501dbfc26a6b6f2dc733da60fa62561c7c157a7b24f69b41313e
+size 37174
diff --git a/records/sd1/sweep/4a0860a9d7c7eaead684/output.webp b/records/sd1/sweep/4a0860a9d7c7eaead684/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..eddfa90d65573cc522dd259ca7a41127aa9972d0
--- /dev/null
+++ b/records/sd1/sweep/4a0860a9d7c7eaead684/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa6ca25f2591ca2fa2699ca44f46357de71058b6be686371aaeebc243b94efce
+size 27224
diff --git a/records/sd1/sweep/4a1026b2930630cea8c5/output.webp b/records/sd1/sweep/4a1026b2930630cea8c5/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..b6b496f4ff47e594815e72ecda52bf1f4cd2bbb6
--- /dev/null
+++ b/records/sd1/sweep/4a1026b2930630cea8c5/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bb7bb8211005e662913d83e542f1104699dc3bc6452539e1ee053ebc832a075
+size 43244
diff --git a/records/sd1/sweep/4a12b67103e25759043a/output.webp b/records/sd1/sweep/4a12b67103e25759043a/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a023078b060cf784513ca0d6317f089b2cb65181
--- /dev/null
+++ b/records/sd1/sweep/4a12b67103e25759043a/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8196e76c86120afbf3e243ecc1edad2029f6f9d1ec4d3e0849fe83ef522791c
+size 56138
diff --git a/records/sd1/sweep/4a17200bf192405081e3/output.webp b/records/sd1/sweep/4a17200bf192405081e3/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..a57b4db682b16d19ab2ea91c669162931a150c9f
--- /dev/null
+++ b/records/sd1/sweep/4a17200bf192405081e3/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea45469ae615504d2ba3ca275d5a108d3eee789aeedb715b4d2dba3918e38f2e
+size 61006
diff --git a/records/sd1/sweep/4a1a57f21a3c242b3b33/output.webp b/records/sd1/sweep/4a1a57f21a3c242b3b33/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..0eec920f292f5f8652d3bc60581a58b08776fe8c
--- /dev/null
+++ b/records/sd1/sweep/4a1a57f21a3c242b3b33/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2fc1516d72cb5c61f07a82b401df1df2b68590daac806d60fe8884317d773456
+size 36498
diff --git a/records/sd1/sweep/4a1bbde3d5566e2a1edf/output.webp b/records/sd1/sweep/4a1bbde3d5566e2a1edf/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..ef40d4d2a61536ca705f4ef3aae940c2d97890ed
--- /dev/null
+++ b/records/sd1/sweep/4a1bbde3d5566e2a1edf/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6912d96251d5c586684de59a578b9396bc0835e785f19dde5dcf14eede51b691
+size 29988
diff --git a/records/sd1/sweep/4a1d4fe059d198a2d859/output.webp b/records/sd1/sweep/4a1d4fe059d198a2d859/output.webp
new file mode 100644
index 0000000000000000000000000000000000000000..c4ad7568719a2fe4f6b22ff9835a3ad8cf60b436
--- /dev/null
+++ b/records/sd1/sweep/4a1d4fe059d198a2d859/output.webp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa2b22fe5325126d540bf6e329a4afd191e30484ad941eb14b7abc034da43641
+size 9392
diff --git a/records/sd1/sweep/4db710a5ae3f7d10177c/record.json b/records/sd1/sweep/4db710a5ae3f7d10177c/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..34604098087a4277aa1aecf569b96fef3ba437f5
--- /dev/null
+++ b/records/sd1/sweep/4db710a5ae3f7d10177c/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4db710a5ae3f7d10177c",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:13Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.2.0.skip_connection",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "out.lo",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/99ac38e17122ee788e3c45a5480baa72bdbcf245ed646c748b0a5abe54180e2e.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "99ac38e17122ee788e3c45a5480baa72bdbcf245ed646c748b0a5abe54180e2e",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "7a5019fd2a49a66e8f98"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4db710a5ae3f7d10177c/workflow.json b/records/sd1/sweep/4db710a5ae3f7d10177c/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..cad7274b3cd6eca195a04e9fe45a7a421b8c7ab1
--- /dev/null
+++ b/records/sd1/sweep/4db710a5ae3f7d10177c/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4db710a5ae3f7d10177c)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.2.0.skip_connection\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dba25cf1969a90da2ae/measurements.json b/records/sd1/sweep/4dba25cf1969a90da2ae/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..8417b1dee1c11fed5441fbfc1e794f1ea3336689
--- /dev/null
+++ b/records/sd1/sweep/4dba25cf1969a90da2ae/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4dba25cf1969a90da2ae",
+ "values": {
+  "cosine_distance": {
+   "value": 0.107263,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.147175,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.044139,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.105077,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 58.921,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00478,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 14.759,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9372,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.442,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dba25cf1969a90da2ae/record.json b/records/sd1/sweep/4dba25cf1969a90da2ae/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..63678826ee400850d63b70b6f94c41b190c5d865
--- /dev/null
+++ b/records/sd1/sweep/4dba25cf1969a90da2ae/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4dba25cf1969a90da2ae",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:34Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.11.0.in_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/35c4cdecbdd2e13ab112d91394729e4a1de8b543cb0864531bf4e6de35d19154.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "35c4cdecbdd2e13ab112d91394729e4a1de8b543cb0864531bf4e6de35d19154",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "a2eb5fae049e92df27d8"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dba25cf1969a90da2ae/workflow.json b/records/sd1/sweep/4dba25cf1969a90da2ae/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a2fcff89547a1b309e44410a98212b8a3ade17f4
--- /dev/null
+++ b/records/sd1/sweep/4dba25cf1969a90da2ae/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4dba25cf1969a90da2ae)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.11.0.in_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dc7c1fe920ad78c7fc8/measurements.json b/records/sd1/sweep/4dc7c1fe920ad78c7fc8/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..48f4cb4b048eb3db0d0732b5032fcd93b153fcc5
--- /dev/null
+++ b/records/sd1/sweep/4dc7c1fe920ad78c7fc8/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4dc7c1fe920ad78c7fc8",
+ "values": {
+  "cosine_distance": {
+   "value": 0.065122,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.081398,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.048878,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.016929,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 57.926,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00618,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 11.896,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0064,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1834,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dc7c1fe920ad78c7fc8/record.json b/records/sd1/sweep/4dc7c1fe920ad78c7fc8/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..297af7254adeb2f07f85af7d4879c71e58579f6f
--- /dev/null
+++ b/records/sd1/sweep/4dc7c1fe920ad78c7fc8/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4dc7c1fe920ad78c7fc8",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:43:46Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.0.0.out_layers.2",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/0005cb090188046cc9c16e098602a86ebf6ab11c1f2e9ee1fb4c113332b7a40a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "0005cb090188046cc9c16e098602a86ebf6ab11c1f2e9ee1fb4c113332b7a40a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "877f3045466699f63d8f"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dc7c1fe920ad78c7fc8/workflow.json b/records/sd1/sweep/4dc7c1fe920ad78c7fc8/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ca4e24fdf1914b3796bad59c1d91b4c84384deae
--- /dev/null
+++ b/records/sd1/sweep/4dc7c1fe920ad78c7fc8/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4dc7c1fe920ad78c7fc8)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.0.0.out_layers.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dcadb9af180f0049446/measurements.json b/records/sd1/sweep/4dcadb9af180f0049446/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..07fb7bcb06e4d31556d03a155feb9ebf5b285872
--- /dev/null
+++ b/records/sd1/sweep/4dcadb9af180f0049446/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4dcadb9af180f0049446",
+ "values": {
+  "cosine_distance": {
+   "value": 0.090171,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.119823,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.086434,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.100622,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 62.896,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00675,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 12.788,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.905,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.51,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dcadb9af180f0049446/record.json b/records/sd1/sweep/4dcadb9af180f0049446/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..c1d5fe6a441a19477863e637e3ccd361993f840b
--- /dev/null
+++ b/records/sd1/sweep/4dcadb9af180f0049446/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4dcadb9af180f0049446",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:17Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.1.proj_out",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.5
+   },
+   "group": "out.mid",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/beac47183ed0655d1a1f97eb47c7c0765d9ba03ee0650caa7953d9292d95ead4.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "beac47183ed0655d1a1f97eb47c7c0765d9ba03ee0650caa7953d9292d95ead4",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "d091cf705a572edac460"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dcadb9af180f0049446/workflow.json b/records/sd1/sweep/4dcadb9af180f0049446/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..76cd8a8f8eb3d2b71bf15ba8fc6b14ae6013f28d
--- /dev/null
+++ b/records/sd1/sweep/4dcadb9af180f0049446/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4dcadb9af180f0049446)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.1.proj_out\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dd8092d77a591f93b50/measurements.json b/records/sd1/sweep/4dd8092d77a591f93b50/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..d2521bbd821b396d4f8d8b494726c80858a558e3
--- /dev/null
+++ b/records/sd1/sweep/4dd8092d77a591f93b50/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4dd8092d77a591f93b50",
+ "values": {
+  "std": {
+   "value": 49.954,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01537,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 8.208,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.04819,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.111878,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.06087,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.019946,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0495,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0304,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dd8092d77a591f93b50/record.json b/records/sd1/sweep/4dd8092d77a591f93b50/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..8a6d7692329902c32dbd7656c293a33f64520e8c
--- /dev/null
+++ b/records/sd1/sweep/4dd8092d77a591f93b50/record.json
@@ -0,0 +1,77 @@
+{
+ "id": "4dd8092d77a591f93b50",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:04:20Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.2,
+    "seed": 7
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "medium"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "13216623a742ccb32c5e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "add_noise",
+   0.2,
+   "all"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "3483aa6a7f8c9372e090"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dd8092d77a591f93b50/workflow.json b/records/sd1/sweep/4dd8092d77a591f93b50/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9092c77a161478c079eae36f0bfe2139cf92743c
--- /dev/null
+++ b/records/sd1/sweep/4dd8092d77a591f93b50/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4dd8092d77a591f93b50)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.2,\n    \"seed\": 7\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dd93fa6c7d014bc651f/interpretations.jsonl b/records/sd1/sweep/4dd93fa6c7d014bc651f/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..88b538f9a3694b94e64996b5db66c325b89c71c6
--- /dev/null
+++ b/records/sd1/sweep/4dd93fa6c7d014bc651f/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "4dd93fa6c7d014bc651f"}, "kind": "caption", "value": "Lighthouse on a rocky island under a blazing red sky, with deep blue glowing water", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:50:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "4b2c40803f2c40a6"}
+{"target": {"record": "4dd93fa6c7d014bc651f"}, "kind": "change", "value": "The palette becomes far more saturated and contrasty, with a fiery red sky, electric blue sea and glowing reflections, giving a dramatic, poster-like look.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:50:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3b0d33ac8b516310"}
+{"target": {"record": "4dd93fa6c7d014bc651f"}, "kind": "keywords", "value": ["lighthouse", "fiery sunset", "vivid colours", "dramatic seascape", "high contrast"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:50:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "83a2168ee578c671"}
+{"target": {"record": "4dd93fa6c7d014bc651f"}, "kind": "effect_tags", "value": ["saturated", "high_contrast", "colour_shift", "poster", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:50:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "6fdb4a09b2a06d5f"}
diff --git a/records/sd1/sweep/4dd93fa6c7d014bc651f/measurements.json b/records/sd1/sweep/4dd93fa6c7d014bc651f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0cd64cd085557c7a59fc881f444798e7dbd3f4f4
--- /dev/null
+++ b/records/sd1/sweep/4dd93fa6c7d014bc651f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4dd93fa6c7d014bc651f",
+ "values": {
+  "std": {
+   "value": 44.589,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.03367,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 32.175,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.15973,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.331632,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.04493,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.049874,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9764,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0416,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dd93fa6c7d014bc651f/record.json b/records/sd1/sweep/4dd93fa6c7d014bc651f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5b23f01938dd25b8c816289b609c9ddc01762af0
--- /dev/null
+++ b/records/sd1/sweep/4dd93fa6c7d014bc651f/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4dd93fa6c7d014bc651f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:15:15Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2.0
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "boost",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "multiply",
+   2.0,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "41dbfac8084e0b3c6232"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4dd93fa6c7d014bc651f/workflow.json b/records/sd1/sweep/4dd93fa6c7d014bc651f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..d47fe975406b11ebbf203690d081033c165e6660
--- /dev/null
+++ b/records/sd1/sweep/4dd93fa6c7d014bc651f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4dd93fa6c7d014bc651f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de47b44fc167c2d0261/measurements.json b/records/sd1/sweep/4de47b44fc167c2d0261/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b02eeaa59cf1f5025b5b158ec94accf12fde30b
--- /dev/null
+++ b/records/sd1/sweep/4de47b44fc167c2d0261/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4de47b44fc167c2d0261",
+ "values": {
+  "cosine_distance": {
+   "value": 0.221451,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.278894,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.07468,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.038712,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 51.444,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00716,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 26.801,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9352,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1102,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de47b44fc167c2d0261/record.json b/records/sd1/sweep/4de47b44fc167c2d0261/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..4f0155d7006c92bf528187db0fb5d7baa6e1d1b4
--- /dev/null
+++ b/records/sd1/sweep/4de47b44fc167c2d0261/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4de47b44fc167c2d0261",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:53Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.11.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/924809aaa3c1547d1ddac381dde3fd25de06903f89b8c6c581cf3c90ed8e9ed4.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "924809aaa3c1547d1ddac381dde3fd25de06903f89b8c6c581cf3c90ed8e9ed4",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "d52f9691a6f40af2010b"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de47b44fc167c2d0261/workflow.json b/records/sd1/sweep/4de47b44fc167c2d0261/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..b8a7b7182091fa58a98772d4b4a97511266fecad
--- /dev/null
+++ b/records/sd1/sweep/4de47b44fc167c2d0261/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4de47b44fc167c2d0261)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.11.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de90c000b8568610238/measurements.json b/records/sd1/sweep/4de90c000b8568610238/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..560849968e2857ba2d2905583b2bbdeed7de5803
--- /dev/null
+++ b/records/sd1/sweep/4de90c000b8568610238/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4de90c000b8568610238",
+ "values": {
+  "std": {
+   "value": 37.396,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01506,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 4.312,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.015745,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.041413,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.023525,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.007793,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0032,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0064,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de90c000b8568610238/record.json b/records/sd1/sweep/4de90c000b8568610238/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..58993827376a939324d91492bd4f09022136335c
--- /dev/null
+++ b/records/sd1/sweep/4de90c000b8568610238/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4de90c000b8568610238",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:14:28Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.8,
+    "seed": 1234
+   },
+   "group": "out.mid",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "heavy",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.mid",
+   "add_noise",
+   0.8,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de90c000b8568610238/workflow.json b/records/sd1/sweep/4de90c000b8568610238/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..afdc061b84c727381d8bb06221cd8ff4de02726b
--- /dev/null
+++ b/records/sd1/sweep/4de90c000b8568610238/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4de90c000b8568610238)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.8,\n    \"seed\": 1234\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de9bbcb9c5c6e06d649/measurements.json b/records/sd1/sweep/4de9bbcb9c5c6e06d649/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..029306880443c5bd76524777bc23da5cd41de91d
--- /dev/null
+++ b/records/sd1/sweep/4de9bbcb9c5c6e06d649/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4de9bbcb9c5c6e06d649",
+ "values": {
+  "std": {
+   "value": 48.828,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01336,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 8.79,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.03337,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.055941,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.008012,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.007969,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0553,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2114,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de9bbcb9c5c6e06d649/record.json b/records/sd1/sweep/4de9bbcb9c5c6e06d649/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..2a5491029678177304f9740591139fbf662817f9
--- /dev/null
+++ b/records/sd1/sweep/4de9bbcb9c5c6e06d649/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4de9bbcb9c5c6e06d649",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:27:05Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.1",
+   "op": "add_scalar",
+   "args": {
+    "scalar": -0.8
+   },
+   "group": "out.lo",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "mid",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "2976534844fe80be26b4",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.lo",
+   "add_scalar",
+   -0.8,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a1719b2aa4f1e46f5f83"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4de9bbcb9c5c6e06d649/workflow.json b/records/sd1/sweep/4de9bbcb9c5c6e06d649/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..5b0536455a9ba316909f0dad7a3dca0e3a7184eb
--- /dev/null
+++ b/records/sd1/sweep/4de9bbcb9c5c6e06d649/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4de9bbcb9c5c6e06d649)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.1\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": -0.8\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4deb18257c7fb58ebe0b/measurements.json b/records/sd1/sweep/4deb18257c7fb58ebe0b/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..fb790e5574322d5b00d3a5fc7e63ad59b53d3cba
--- /dev/null
+++ b/records/sd1/sweep/4deb18257c7fb58ebe0b/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4deb18257c7fb58ebe0b",
+ "values": {
+  "cosine_distance": {
+   "value": 0.224121,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.350215,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.197016,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.093289,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 50.426,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00386,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 29.086,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9571,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0364,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4deb18257c7fb58ebe0b/record.json b/records/sd1/sweep/4deb18257c7fb58ebe0b/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a280aa8b78792bdfc5c9586e483b1b5069fbf1f7
--- /dev/null
+++ b/records/sd1/sweep/4deb18257c7fb58ebe0b/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4deb18257c7fb58ebe0b",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:41:07Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.4.1.transformer_blocks.0.ff.net.2",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "in.mid",
+   "kind": "ff",
+   "module_type": "Linear",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/16fae9b948296a76a914274a197f37c6701a70cf7ddd3a87e36aac9c73390fb4.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "16fae9b948296a76a914274a197f37c6701a70cf7ddd3a87e36aac9c73390fb4",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.FeedForward.Sequential.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "cc4c47b27aa096929379"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4deb18257c7fb58ebe0b/workflow.json b/records/sd1/sweep/4deb18257c7fb58ebe0b/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..61fbaae4adc61a71cb0368e1661d3ef18206b7c4
--- /dev/null
+++ b/records/sd1/sweep/4deb18257c7fb58ebe0b/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4deb18257c7fb58ebe0b)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.4.1.transformer_blocks.0.ff.net.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4df0452e3e55dbf1da58/measurements.json b/records/sd1/sweep/4df0452e3e55dbf1da58/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..52aac18cd53498ced65d8bf18e0040380266ea6e
--- /dev/null
+++ b/records/sd1/sweep/4df0452e3e55dbf1da58/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4df0452e3e55dbf1da58",
+ "values": {
+  "cosine_distance": {
+   "value": 0.142711,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.185573,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.053204,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.039107,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.978,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00549,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 18.835,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9931,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.075,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4df0452e3e55dbf1da58/record.json b/records/sd1/sweep/4df0452e3e55dbf1da58/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b06c87cccfa27e96baae88054cdf0e738e27d9ff
--- /dev/null
+++ b/records/sd1/sweep/4df0452e3e55dbf1da58/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4df0452e3e55dbf1da58",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:41:45Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.1.transformer_blocks.0.norm2",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.mid",
+   "kind": "norm",
+   "module_type": "LayerNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/83bb528a7f59070130b568d3f4be91b8fcbd032b55fb774fbe5fe787f2c4d3c5.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "83bb528a7f59070130b568d3f4be91b8fcbd032b55fb774fbe5fe787f2c4d3c5",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.LayerNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "f25c80d340913ca758bf"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4df0452e3e55dbf1da58/workflow.json b/records/sd1/sweep/4df0452e3e55dbf1da58/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..e2ed4a2fc177319cfc7e798fe21b6d6eceebb8a3
--- /dev/null
+++ b/records/sd1/sweep/4df0452e3e55dbf1da58/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4df0452e3e55dbf1da58)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.1.transformer_blocks.0.norm2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e009b4bcb84966d35d2/measurements.json b/records/sd1/sweep/4e009b4bcb84966d35d2/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..80a9ca0558bcb0791069bd56ecf8215525a69766
--- /dev/null
+++ b/records/sd1/sweep/4e009b4bcb84966d35d2/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e009b4bcb84966d35d2",
+ "values": {
+  "std": {
+   "value": 42.12,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01466,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 17.405,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.119653,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.348475,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.448308,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.153999,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9799,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0836,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e009b4bcb84966d35d2/record.json b/records/sd1/sweep/4e009b4bcb84966d35d2/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d0baafd403aac230bf7cff15f5c07124a4010bf4
--- /dev/null
+++ b/records/sd1/sweep/4e009b4bcb84966d35d2/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4e009b4bcb84966d35d2",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:49:53Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.7
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "shrink",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "scale",
+   0.7,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "7c1322f9d1f44b290d10"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e009b4bcb84966d35d2/workflow.json b/records/sd1/sweep/4e009b4bcb84966d35d2/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..bc153303a91839806c9a6aa3c7b5978865dfd43b
--- /dev/null
+++ b/records/sd1/sweep/4e009b4bcb84966d35d2/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e009b4bcb84966d35d2)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.7\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e10186b41e704ac00d4/measurements.json b/records/sd1/sweep/4e10186b41e704ac00d4/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..ede7981f1166668df6bdd373703e6b1064e57ea6
--- /dev/null
+++ b/records/sd1/sweep/4e10186b41e704ac00d4/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4e10186b41e704ac00d4",
+ "values": {
+  "cosine_distance": {
+   "value": 0.774064,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.713953,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.938195,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.566848,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 57.999,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01928,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": true,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": [
+    "clip_degenerate"
+   ]
+  },
+  "mae_vs_baseline": {
+   "value": 77.209,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.6064,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.9317,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e10186b41e704ac00d4/record.json b/records/sd1/sweep/4e10186b41e704ac00d4/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a37da9744b91ee8d3691543625b285225a177d68
--- /dev/null
+++ b/records/sd1/sweep/4e10186b41e704ac00d4/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4e10186b41e704ac00d4",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:43Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.0.skip_connection",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "out.mid",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/21da50778f1ec2bf090dc76d41665b04c8cf89ab0d1d6aa83ec98fa8d82aba10.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "21da50778f1ec2bf090dc76d41665b04c8cf89ab0d1d6aa83ec98fa8d82aba10",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "e12f8235b891f7a6d6ef"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e10186b41e704ac00d4/workflow.json b/records/sd1/sweep/4e10186b41e704ac00d4/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c4fd5f1ee3458a831c779262a4c4d7a268168bf7
--- /dev/null
+++ b/records/sd1/sweep/4e10186b41e704ac00d4/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e10186b41e704ac00d4)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.0.skip_connection\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e12c83598612980bedf/measurements.json b/records/sd1/sweep/4e12c83598612980bedf/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..eb4f6e5fc0fa9718781741a07978de8b59a91005
--- /dev/null
+++ b/records/sd1/sweep/4e12c83598612980bedf/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e12c83598612980bedf",
+ "values": {
+  "std": {
+   "value": 51.456,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01055,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 4.92,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.013463,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.045014,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.022818,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.006185,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9912,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0124,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e12c83598612980bedf/record.json b/records/sd1/sweep/4e12c83598612980bedf/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9bbfd2e2c9840d179a20b3fa89970d82e7d125b9
--- /dev/null
+++ b/records/sd1/sweep/4e12c83598612980bedf/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4e12c83598612980bedf",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:35:56Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "mid",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "add_scalar",
+   0.8,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a03bea9625a470c85e04"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e12c83598612980bedf/workflow.json b/records/sd1/sweep/4e12c83598612980bedf/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..3a1462550649e21a370c0895563d165c3f7df512
--- /dev/null
+++ b/records/sd1/sweep/4e12c83598612980bedf/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e12c83598612980bedf)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e170a34016d234d9a1e/measurements.json b/records/sd1/sweep/4e170a34016d234d9a1e/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..18130b1ac849e106b7961f86012ab6201185f413
--- /dev/null
+++ b/records/sd1/sweep/4e170a34016d234d9a1e/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e170a34016d234d9a1e",
+ "values": {
+  "std": {
+   "value": 68.845,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00719,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 63.811,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.622417,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.699061,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.853107,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.392002,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.6739,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.6917,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e170a34016d234d9a1e/record.json b/records/sd1/sweep/4e170a34016d234d9a1e/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5a5673a6d02336f05b8cc4ac71f42e34acbe932f
--- /dev/null
+++ b/records/sd1/sweep/4e170a34016d234d9a1e/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "4e170a34016d234d9a1e",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:23:10Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90.0
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "quarter",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "13216623a742ccb32c5e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "rotate",
+   90.0,
+   "middle"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 90.0,
+    "applied": 90.0,
+    "reason": "outside safe range [0.0, 5.0] for out.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "902ffa6e0f6f24935832"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e170a34016d234d9a1e/workflow.json b/records/sd1/sweep/4e170a34016d234d9a1e/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..00646124a8191afc1d7f43dde6945f5612d01628
--- /dev/null
+++ b/records/sd1/sweep/4e170a34016d234d9a1e/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e170a34016d234d9a1e)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e1f6cb9f5b058c15e1b/measurements.json b/records/sd1/sweep/4e1f6cb9f5b058c15e1b/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..bd144fb21542c35e90cf2e11fb7a7c9155dc9df2
--- /dev/null
+++ b/records/sd1/sweep/4e1f6cb9f5b058c15e1b/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e1f6cb9f5b058c15e1b",
+ "values": {
+  "cosine_distance": {
+   "value": 0.234668,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.303439,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.115885,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.059694,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.216,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00602,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 31.124,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9719,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1068,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e1f6cb9f5b058c15e1b/record.json b/records/sd1/sweep/4e1f6cb9f5b058c15e1b/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a76cb17cd5a623f25c133445e3f29ee0b3088da2
--- /dev/null
+++ b/records/sd1/sweep/4e1f6cb9f5b058c15e1b/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4e1f6cb9f5b058c15e1b",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:42:37Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.1.transformer_blocks.0.attn2.to_out.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.lo",
+   "kind": "cross_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/622082af13370ec2b11899aa6b97061e45958e516a03b7de54cc4d368895733c.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "622082af13370ec2b11899aa6b97061e45958e516a03b7de54cc4d368895733c",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Sequential.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6e72519422536cc81c45"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e1f6cb9f5b058c15e1b/workflow.json b/records/sd1/sweep/4e1f6cb9f5b058c15e1b/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..df3d5ddca77ef6fab51d59c66137505d619658eb
--- /dev/null
+++ b/records/sd1/sweep/4e1f6cb9f5b058c15e1b/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e1f6cb9f5b058c15e1b)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.1.transformer_blocks.0.attn2.to_out.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e24c2c7ad5a888e3c8e/measurements.json b/records/sd1/sweep/4e24c2c7ad5a888e3c8e/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..83e782d6a8435e97ffa11b6074af83009d8a6db4
--- /dev/null
+++ b/records/sd1/sweep/4e24c2c7ad5a888e3c8e/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e24c2c7ad5a888e3c8e",
+ "values": {
+  "cosine_distance": {
+   "value": 0.203389,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.371502,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.654593,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.296778,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 49.015,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00455,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.806,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0564,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7556,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e24c2c7ad5a888e3c8e/record.json b/records/sd1/sweep/4e24c2c7ad5a888e3c8e/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..efaf1f0aa2e39691bdd54963ddcad7361c3674ee
--- /dev/null
+++ b/records/sd1/sweep/4e24c2c7ad5a888e3c8e/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4e24c2c7ad5a888e3c8e",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:15Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0.in_layers.2",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.5
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/9284bbc6567470f2baddc96a36db4ff4ec710601524b9b116eaef31dee17a403.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "9284bbc6567470f2baddc96a36db4ff4ec710601524b9b116eaef31dee17a403",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "e9e964a277ea4750a8e3"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e24c2c7ad5a888e3c8e/workflow.json b/records/sd1/sweep/4e24c2c7ad5a888e3c8e/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..69bf74093e5eda94d43e908db9527b7fd7292b38
--- /dev/null
+++ b/records/sd1/sweep/4e24c2c7ad5a888e3c8e/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e24c2c7ad5a888e3c8e)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0.in_layers.2\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e25774f16c36b2e4393/measurements.json b/records/sd1/sweep/4e25774f16c36b2e4393/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..3fa2e7633c5f519cbd7fbf1d3e2f27db8e48d847
--- /dev/null
+++ b/records/sd1/sweep/4e25774f16c36b2e4393/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4e25774f16c36b2e4393",
+ "values": {
+  "cosine_distance": {
+   "value": 0.159936,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.262109,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.298103,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.26866,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.177,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00381,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 20.068,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9476,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2935,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e25774f16c36b2e4393/record.json b/records/sd1/sweep/4e25774f16c36b2e4393/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..66a6f6165994277ce932621757d4e9383d463f33
--- /dev/null
+++ b/records/sd1/sweep/4e25774f16c36b2e4393/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4e25774f16c36b2e4393",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:17Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.in_layers.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.5
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/7fbc34b603b6cef5fdf545f598378fcf87a0d11bc3625959a8d7588a06089fbf.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "7fbc34b603b6cef5fdf545f598378fcf87a0d11bc3625959a8d7588a06089fbf",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "66ba4c2819f062e7cbe4"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e25774f16c36b2e4393/workflow.json b/records/sd1/sweep/4e25774f16c36b2e4393/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a156fc2a8f70e823c92154f14737a0f49555a65d
--- /dev/null
+++ b/records/sd1/sweep/4e25774f16c36b2e4393/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e25774f16c36b2e4393)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.in_layers.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e298ee8b0a936bf282a/measurements.json b/records/sd1/sweep/4e298ee8b0a936bf282a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..4f378352d8cbd8f9e3bd13ba44d801a59dfc4d5c
--- /dev/null
+++ b/records/sd1/sweep/4e298ee8b0a936bf282a/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e298ee8b0a936bf282a",
+ "values": {
+  "std": {
+   "value": 57.417,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01075,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 19.164,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.088065,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.115077,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.042316,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.009541,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0009,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2011,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e298ee8b0a936bf282a/record.json b/records/sd1/sweep/4e298ee8b0a936bf282a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..48bcd59c028f560d6e6fc4c3344ac57c81c959e0
--- /dev/null
+++ b/records/sd1/sweep/4e298ee8b0a936bf282a/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4e298ee8b0a936bf282a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T21:57:39Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "mid",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "add_scalar",
+   0.8,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "516e49415018ab80811b"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e298ee8b0a936bf282a/workflow.json b/records/sd1/sweep/4e298ee8b0a936bf282a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a5e72ae6afde2dbd75a3fcc4f647ff18b5e19261
--- /dev/null
+++ b/records/sd1/sweep/4e298ee8b0a936bf282a/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e298ee8b0a936bf282a)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e2caaf41239fd586f38/measurements.json b/records/sd1/sweep/4e2caaf41239fd586f38/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..01bfd5365ccb9d7671d2da314786b2828a5f4b5e
--- /dev/null
+++ b/records/sd1/sweep/4e2caaf41239fd586f38/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4e2caaf41239fd586f38",
+ "values": {
+  "cosine_distance": {
+   "value": 0.575211,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.697237,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.483139,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.22347,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.994,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00271,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 54.857,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.979,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4983,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e2caaf41239fd586f38/record.json b/records/sd1/sweep/4e2caaf41239fd586f38/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a9d0b786eeca5c056a2443a488e9e8ad98cc8762
--- /dev/null
+++ b/records/sd1/sweep/4e2caaf41239fd586f38/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4e2caaf41239fd586f38",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:00Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.0.0.out_layers.3",
+   "op": "scale",
+   "args": {
+    "scale_factor": 2
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "strong_grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/d23e3ddfcd1d03eaada163230b8d32b5ef0ffbc092e14f3a73b5a77d95161568.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "d23e3ddfcd1d03eaada163230b8d32b5ef0ffbc092e14f3a73b5a77d95161568",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "52020b525ce7692b75f3"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e2caaf41239fd586f38/workflow.json b/records/sd1/sweep/4e2caaf41239fd586f38/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a000d46af02b4cf2513cf691d7c6761039f9a95b
--- /dev/null
+++ b/records/sd1/sweep/4e2caaf41239fd586f38/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e2caaf41239fd586f38)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.0.0.out_layers.3\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e37653487d13ae7c832/measurements.json b/records/sd1/sweep/4e37653487d13ae7c832/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..88365edb287e42e5f61dc6f54d1f0ca88c5e4b83
--- /dev/null
+++ b/records/sd1/sweep/4e37653487d13ae7c832/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e37653487d13ae7c832",
+ "values": {
+  "std": {
+   "value": 41.907,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00831,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 32.219,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.329962,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.516581,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.173281,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.071126,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9835,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0537,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e37653487d13ae7c832/record.json b/records/sd1/sweep/4e37653487d13ae7c832/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..e02d25d2442ba40ee4d03641d031c0c965408e68
--- /dev/null
+++ b/records/sd1/sweep/4e37653487d13ae7c832/record.json
@@ -0,0 +1,84 @@
+{
+ "id": "4e37653487d13ae7c832",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:48:33Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": -0.8
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "mid",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "add_scalar",
+   -0.8,
+   "middle"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": -0.8,
+    "applied": -0.8,
+    "reason": "outside safe range [-0.5, 1.5] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e37653487d13ae7c832/workflow.json b/records/sd1/sweep/4e37653487d13ae7c832/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7132cc2b35f85ae922f7d1956febd9ed312c875c
--- /dev/null
+++ b/records/sd1/sweep/4e37653487d13ae7c832/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e37653487d13ae7c832)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": -0.8\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e3d92c95c66ce92e859/measurements.json b/records/sd1/sweep/4e3d92c95c66ce92e859/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ea3cfe575502294aba7fe68ef8b2e943f6c1811
--- /dev/null
+++ b/records/sd1/sweep/4e3d92c95c66ce92e859/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e3d92c95c66ce92e859",
+ "values": {
+  "cosine_distance": {
+   "value": 0.234172,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.269286,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.301243,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.087196,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.856,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00806,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.775,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0179,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1922,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e3d92c95c66ce92e859/record.json b/records/sd1/sweep/4e3d92c95c66ce92e859/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5f715d6d8e54c4bfd3d98a69a3e7a5958ff4756b
--- /dev/null
+++ b/records/sd1/sweep/4e3d92c95c66ce92e859/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4e3d92c95c66ce92e859",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:44:56Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.4.1.transformer_blocks.0.norm2",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "out.lo",
+   "kind": "norm",
+   "module_type": "LayerNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/bab378489dd6f38e3256570b6d2a59e30d9cc32128ac03cd4f9f6caef52bb007.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "bab378489dd6f38e3256570b6d2a59e30d9cc32128ac03cd4f9f6caef52bb007",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.LayerNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "4a76d52dea3abbc5c115"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e3d92c95c66ce92e859/workflow.json b/records/sd1/sweep/4e3d92c95c66ce92e859/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..62bb297c89d818f6885d6b1275fdf21c6fab45ae
--- /dev/null
+++ b/records/sd1/sweep/4e3d92c95c66ce92e859/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e3d92c95c66ce92e859)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.4.1.transformer_blocks.0.norm2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e47d4de56a261169936/measurements.json b/records/sd1/sweep/4e47d4de56a261169936/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..b591791298d2d92c81330617d4abb1dd4f8b950c
--- /dev/null
+++ b/records/sd1/sweep/4e47d4de56a261169936/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4e47d4de56a261169936",
+ "values": {
+  "cosine_distance": {
+   "value": 0.08581,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.159499,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.12236,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.08628,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.568,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00359,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 14.568,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0327,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3597,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e47d4de56a261169936/record.json b/records/sd1/sweep/4e47d4de56a261169936/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..4653b800b6741901bc6fcbe5d0f69bced98a8cab
--- /dev/null
+++ b/records/sd1/sweep/4e47d4de56a261169936/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4e47d4de56a261169936",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:18Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.0.out_layers.3",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/6eb1d1b7159da294bfcaedf256ae5e8bbb2c16e5d5afac051f12168984ad7106.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "6eb1d1b7159da294bfcaedf256ae5e8bbb2c16e5d5afac051f12168984ad7106",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "0a6e0ddda1686df335e9"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e47d4de56a261169936/workflow.json b/records/sd1/sweep/4e47d4de56a261169936/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..680d7f779e29dc8a004f7aae6f61c093676a072a
--- /dev/null
+++ b/records/sd1/sweep/4e47d4de56a261169936/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e47d4de56a261169936)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.0.out_layers.3\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e60be25896e21d11bc4/measurements.json b/records/sd1/sweep/4e60be25896e21d11bc4/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..eb89e58121f8f1fcdd305f29740b430e10453a51
--- /dev/null
+++ b/records/sd1/sweep/4e60be25896e21d11bc4/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4e60be25896e21d11bc4",
+ "values": {
+  "cosine_distance": {
+   "value": 0.0,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.0,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.618,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00534,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.0,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1553,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e60be25896e21d11bc4/record.json b/records/sd1/sweep/4e60be25896e21d11bc4/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..8eef22e7fd42baaf8dda9c9688d94149cc03eec1
--- /dev/null
+++ b/records/sd1/sweep/4e60be25896e21d11bc4/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4e60be25896e21d11bc4",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:49:22Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.1.0.h_upd",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/365752f7784ff53e4272c30787d6693b478d060fe9206dbadeac3fe55825525b.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "365752f7784ff53e4272c30787d6693b478d060fe9206dbadeac3fe55825525b",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "17be94f5d8c224cbc123"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e60be25896e21d11bc4/workflow.json b/records/sd1/sweep/4e60be25896e21d11bc4/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..d1529d29ffe5e657af097cd2d0638dc63dda04ed
--- /dev/null
+++ b/records/sd1/sweep/4e60be25896e21d11bc4/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e60be25896e21d11bc4)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.1.0.h_upd\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e6c6f4998239f858d8c/interpretations.jsonl b/records/sd1/sweep/4e6c6f4998239f858d8c/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..9f300c3b086f5666bcb4c8dab84abd4b53c68c6f
--- /dev/null
+++ b/records/sd1/sweep/4e6c6f4998239f858d8c/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "4e6c6f4998239f858d8c"}, "kind": "caption", "value": "Soft painted woman with dark brown bob against a muted grey-green background", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:21:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ecd7c110406db8a3"}
+{"target": {"record": "4e6c6f4998239f858d8c"}, "kind": "change", "value": "The face softens and rounds, the hair turns warmer dark brown with looser waves, and the background becomes a plain muted grey-green.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:21:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "c203405adddc40b0"}
+{"target": {"record": "4e6c6f4998239f858d8c"}, "kind": "keywords", "value": ["soft portrait", "dark brown bob", "muted background"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:21:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "0aa130b2455cf4c0"}
+{"target": {"record": "4e6c6f4998239f858d8c"}, "kind": "effect_tags", "value": ["subject_kept", "painterly", "colour_shift"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:21:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "0f46b34edfefca24"}
diff --git a/records/sd1/sweep/4e6c6f4998239f858d8c/measurements.json b/records/sd1/sweep/4e6c6f4998239f858d8c/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..670ea7a0961c523f336f680f11d07ab827cb41b8
--- /dev/null
+++ b/records/sd1/sweep/4e6c6f4998239f858d8c/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4e6c6f4998239f858d8c",
+ "values": {
+  "cosine_distance": {
+   "value": 0.248195,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.332553,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.184437,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.062493,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.86,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.006,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 29.869,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9719,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0514,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e6c6f4998239f858d8c/record.json b/records/sd1/sweep/4e6c6f4998239f858d8c/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ddd59c5ce41c3bc39d6cccfc068949a3c2b77fce
--- /dev/null
+++ b/records/sd1/sweep/4e6c6f4998239f858d8c/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4e6c6f4998239f858d8c",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:18Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "in.mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/6c46d425b6c62794fe1962232838eecc6b50d12ceace389c1043b470ab70b6d6.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "6c46d425b6c62794fe1962232838eecc6b50d12ceace389c1043b470ab70b6d6",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "5d9b8d3869ec71265633"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e6c6f4998239f858d8c/workflow.json b/records/sd1/sweep/4e6c6f4998239f858d8c/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..464d1bf3b0d798082c11f0c6be15755bac4fa99e
--- /dev/null
+++ b/records/sd1/sweep/4e6c6f4998239f858d8c/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e6c6f4998239f858d8c)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e8bd3175223bd6d44ed/measurements.json b/records/sd1/sweep/4e8bd3175223bd6d44ed/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..53ef2a1f6d85ae018bdf196c5f65f630697cb2af
--- /dev/null
+++ b/records/sd1/sweep/4e8bd3175223bd6d44ed/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4e8bd3175223bd6d44ed",
+ "values": {
+  "cosine_distance": {
+   "value": 0.263908,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.331541,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.202127,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.120923,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.796,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00459,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 29.347,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9729,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2141,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e8bd3175223bd6d44ed/record.json b/records/sd1/sweep/4e8bd3175223bd6d44ed/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..096adf2f29d1e94221c3601a3ed94cda24102cce
--- /dev/null
+++ b/records/sd1/sweep/4e8bd3175223bd6d44ed/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4e8bd3175223bd6d44ed",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:42:27Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.1.proj_in",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "in.lo",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/82efbf6ba245be780845a204570d115ef809c3da353ccc3860c19b993487be01.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "82efbf6ba245be780845a204570d115ef809c3da353ccc3860c19b993487be01",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "52d60d49744c6273d08e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e8bd3175223bd6d44ed/workflow.json b/records/sd1/sweep/4e8bd3175223bd6d44ed/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..22091849c57c70decf663daffafade061e2adf58
--- /dev/null
+++ b/records/sd1/sweep/4e8bd3175223bd6d44ed/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e8bd3175223bd6d44ed)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.1.proj_in\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e8f21c73c7d4e27f3bc/measurements.json b/records/sd1/sweep/4e8f21c73c7d4e27f3bc/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..1b4af160d34ac981ac644717df1b37907baa2042
--- /dev/null
+++ b/records/sd1/sweep/4e8f21c73c7d4e27f3bc/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4e8f21c73c7d4e27f3bc",
+ "values": {
+  "cosine_distance": {
+   "value": 0.186057,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.321768,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.175003,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.140172,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 54.413,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00567,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 24.834,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.989,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0234,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e8f21c73c7d4e27f3bc/record.json b/records/sd1/sweep/4e8f21c73c7d4e27f3bc/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9ba0bbc6a3fe7e5667c28f93bc55d794ef5551a0
--- /dev/null
+++ b/records/sd1/sweep/4e8f21c73c7d4e27f3bc/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4e8f21c73c7d4e27f3bc",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:50:29Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.0.out_layers.3",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/5ed0fc953c027ba40515feb38aff49db438b4b83726001ac4baf2a77e3ee2def.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "5ed0fc953c027ba40515feb38aff49db438b4b83726001ac4baf2a77e3ee2def",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "03e4c36a95b8da9aea45"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4e8f21c73c7d4e27f3bc/workflow.json b/records/sd1/sweep/4e8f21c73c7d4e27f3bc/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..3f8f8a29fd688d9db052492678b6f52a2a6fc720
--- /dev/null
+++ b/records/sd1/sweep/4e8f21c73c7d4e27f3bc/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4e8f21c73c7d4e27f3bc)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.0.out_layers.3\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea13349aadc0fd436cf/measurements.json b/records/sd1/sweep/4ea13349aadc0fd436cf/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0b9846c9bc8c5b21033503fc94ddabe023c40369
--- /dev/null
+++ b/records/sd1/sweep/4ea13349aadc0fd436cf/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4ea13349aadc0fd436cf",
+ "values": {
+  "cosine_distance": {
+   "value": 0.143009,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.202814,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.078848,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.02796,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 49.377,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00533,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 20.729,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9409,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1267,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea13349aadc0fd436cf/record.json b/records/sd1/sweep/4ea13349aadc0fd436cf/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d2d99e2e19975f00c336e878c7d23188371751b1
--- /dev/null
+++ b/records/sd1/sweep/4ea13349aadc0fd436cf/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4ea13349aadc0fd436cf",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:29Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0.emb_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/060778363d2f495ba887fe882e116876d19025b776023ac892a815c3d0bcf8d8.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "060778363d2f495ba887fe882e116876d19025b776023ac892a815c3d0bcf8d8",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "340ab11556e081f4c992"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea13349aadc0fd436cf/workflow.json b/records/sd1/sweep/4ea13349aadc0fd436cf/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d6587d9d0b1a1129b70aa29e8012723bde41331
--- /dev/null
+++ b/records/sd1/sweep/4ea13349aadc0fd436cf/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4ea13349aadc0fd436cf)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0.emb_layers.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea1fe480abfd7a1966d/measurements.json b/records/sd1/sweep/4ea1fe480abfd7a1966d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..3795404c6835e5bdf2436880fa4aa1a585fcd080
--- /dev/null
+++ b/records/sd1/sweep/4ea1fe480abfd7a1966d/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4ea1fe480abfd7a1966d",
+ "values": {
+  "std": {
+   "value": 43.299,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01674,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 18.108,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.145195,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.29207,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.091967,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.022237,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0199,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0187,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea1fe480abfd7a1966d/record.json b/records/sd1/sweep/4ea1fe480abfd7a1966d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..22596bbebba407600debda9ab8262514e4d6b230
--- /dev/null
+++ b/records/sd1/sweep/4ea1fe480abfd7a1966d/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4ea1fe480abfd7a1966d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:50:20Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.8,
+    "seed": 7
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "heavy",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "add_noise",
+   0.8,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea1fe480abfd7a1966d/workflow.json b/records/sd1/sweep/4ea1fe480abfd7a1966d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..e7396b08785202039fd9b538d50e6a545c33bebe
--- /dev/null
+++ b/records/sd1/sweep/4ea1fe480abfd7a1966d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4ea1fe480abfd7a1966d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.8,\n    \"seed\": 7\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea85efb20527a80dbe6/measurements.json b/records/sd1/sweep/4ea85efb20527a80dbe6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..ed74216524144c9d64c3789e4d512f2c7e28a634
--- /dev/null
+++ b/records/sd1/sweep/4ea85efb20527a80dbe6/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4ea85efb20527a80dbe6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.183886,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.244793,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.147053,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.231833,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.615,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00674,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 19.909,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8961,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3656,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea85efb20527a80dbe6/record.json b/records/sd1/sweep/4ea85efb20527a80dbe6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..776c638988c6422841f1fbf164f4be464ceea929
--- /dev/null
+++ b/records/sd1/sweep/4ea85efb20527a80dbe6/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4ea85efb20527a80dbe6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:01Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.5.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/ad6f8dbbb12e17ab893a11ddc7d0f2c2e0943f7a105766356db026e8ee7ce584.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "ad6f8dbbb12e17ab893a11ddc7d0f2c2e0943f7a105766356db026e8ee7ce584",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "3e1307e58b155e401c82"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ea85efb20527a80dbe6/workflow.json b/records/sd1/sweep/4ea85efb20527a80dbe6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..58841596f3298938559ca7e04320178a272c6292
--- /dev/null
+++ b/records/sd1/sweep/4ea85efb20527a80dbe6/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4ea85efb20527a80dbe6)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.5.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eaba1c7630b8f6c462f/interpretations.jsonl b/records/sd1/sweep/4eaba1c7630b8f6c462f/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..ffd5c72f6125c8cedaaae7e015e203d161a64b8d
--- /dev/null
+++ b/records/sd1/sweep/4eaba1c7630b8f6c462f/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "4eaba1c7630b8f6c462f"}, "kind": "caption", "value": "Grid of glitchy blue and red rectangular panels resembling a fragmented building facade", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:57:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "8cc39739fed83c13"}
+{"target": {"record": "4eaba1c7630b8f6c462f"}, "kind": "change", "value": "The street turns into a flat mosaic of blue and red rectangles in a grid, like a pixelated skyscraper facade, losing all perspective and the neon signs.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:57:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a2b7dd789d4a35cb"}
+{"target": {"record": "4eaba1c7630b8f6c462f"}, "kind": "keywords", "value": ["grid mosaic", "glitch facade", "pixel blocks", "blue panels", "fragmented city"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:57:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "0f2574f10b4b8223"}
+{"target": {"record": "4eaba1c7630b8f6c462f"}, "kind": "effect_tags", "value": ["geometric", "glitch", "pattern", "fragmented", "subject_lost", "abstract"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:57:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "781021b7f68bdb04"}
diff --git a/records/sd1/sweep/4eaba1c7630b8f6c462f/measurements.json b/records/sd1/sweep/4eaba1c7630b8f6c462f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..1b40ee19af9a787fab8a25dd106d1610cd4945c9
--- /dev/null
+++ b/records/sd1/sweep/4eaba1c7630b8f6c462f/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4eaba1c7630b8f6c462f",
+ "values": {
+  "std": {
+   "value": 26.269,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.04001,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": true,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": [
+    "clip_degenerate"
+   ]
+  },
+  "mae_vs_baseline": {
+   "value": 45.903,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.401631,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.59398,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.851564,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.261889,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.7099,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.9076,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eaba1c7630b8f6c462f/record.json b/records/sd1/sweep/4eaba1c7630b8f6c462f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..6591ff3e046fa404074969ca9f517a4474e07377
--- /dev/null
+++ b/records/sd1/sweep/4eaba1c7630b8f6c462f/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "4eaba1c7630b8f6c462f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:17:54Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "multiply",
+   0.0,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for in.lo (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "b1ca4a166e0ddab24277"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eaba1c7630b8f6c462f/workflow.json b/records/sd1/sweep/4eaba1c7630b8f6c462f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..869ce06699f1b0d636757dc8dfc6d890ce872720
--- /dev/null
+++ b/records/sd1/sweep/4eaba1c7630b8f6c462f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4eaba1c7630b8f6c462f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eae94effd70016f49a6/interpretations.jsonl b/records/sd1/sweep/4eae94effd70016f49a6/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c87e33d5fa5ec2ef44c5fb08dd82d4b7ca154fd8
--- /dev/null
+++ b/records/sd1/sweep/4eae94effd70016f49a6/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "4eae94effd70016f49a6"}, "kind": "caption", "value": "Small framed painting of a sunset lighthouse hanging on a dark brown wall", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:48:27Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "395a3a8ddeb3fdb1"}
+{"target": {"record": "4eae94effd70016f49a6"}, "kind": "change", "value": "The full-frame seascape is reduced to a smaller picture within a picture, placed as a framed canvas on a dark textured wall.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:48:27Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "d76cd0dd2897f27f"}
+{"target": {"record": "4eae94effd70016f49a6"}, "kind": "keywords", "value": ["picture in picture", "framed canvas", "lighthouse", "dark wall", "shrunken scene"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:48:27Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ea31cb15dd0bdeb7"}
+{"target": {"record": "4eae94effd70016f49a6"}, "kind": "effect_tags", "value": ["framing", "subject_kept", "dark", "new_composition"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:48:27Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3344a926eb28b934"}
diff --git a/records/sd1/sweep/4eae94effd70016f49a6/measurements.json b/records/sd1/sweep/4eae94effd70016f49a6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..5d9182f097d45279a3cb78931f8ced313d7cb134
--- /dev/null
+++ b/records/sd1/sweep/4eae94effd70016f49a6/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4eae94effd70016f49a6",
+ "values": {
+  "std": {
+   "value": 38.73,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01139,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 33.556,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.430671,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.650626,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.260061,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.174287,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0014,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0525,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eae94effd70016f49a6/record.json b/records/sd1/sweep/4eae94effd70016f49a6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..44bea82c19a7ddce0d89836618bc7419cc2d87e8
--- /dev/null
+++ b/records/sd1/sweep/4eae94effd70016f49a6/record.json
@@ -0,0 +1,77 @@
+{
+ "id": "4eae94effd70016f49a6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:49:51Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.7
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "shrink",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "scale",
+   0.7,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eae94effd70016f49a6/workflow.json b/records/sd1/sweep/4eae94effd70016f49a6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ae1322aa16688d2f772d07d17f94701f0a174b1d
--- /dev/null
+++ b/records/sd1/sweep/4eae94effd70016f49a6/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4eae94effd70016f49a6)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.7\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb2750ea19a1c846014/interpretations.jsonl b/records/sd1/sweep/4eb2750ea19a1c846014/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c79e601826d2146320b294945dabee0126bd1607
--- /dev/null
+++ b/records/sd1/sweep/4eb2750ea19a1c846014/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "4eb2750ea19a1c846014"}, "kind": "caption", "value": "Painted portrait of a woman with a brown bob and black turtleneck on grey background", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:15:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "abd71bc8e45be8c5"}
+{"target": {"record": "4eb2750ea19a1c846014"}, "kind": "change", "value": "A modest variation: the hair becomes a shorter, smoother brown bob, the face is slightly redrawn, the background is a lighter grey, and the border is removed.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:15:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "6d1cba8d7f1dffa2"}
+{"target": {"record": "4eb2750ea19a1c846014"}, "kind": "keywords", "value": ["portrait", "brown bob", "grey background", "variation"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:15:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "20f2ec5746ea9f70"}
+{"target": {"record": "4eb2750ea19a1c846014"}, "kind": "effect_tags", "value": ["subtle", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:15:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "7b074532f3eef24f"}
diff --git a/records/sd1/sweep/4eb2750ea19a1c846014/measurements.json b/records/sd1/sweep/4eb2750ea19a1c846014/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c8a96856f7ca80e97dd40877c6dd9e6881177068
--- /dev/null
+++ b/records/sd1/sweep/4eb2750ea19a1c846014/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4eb2750ea19a1c846014",
+ "values": {
+  "cosine_distance": {
+   "value": 0.212923,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.321371,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.168473,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.081893,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.795,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00477,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 27.844,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0042,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0425,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb2750ea19a1c846014/record.json b/records/sd1/sweep/4eb2750ea19a1c846014/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3d3431896182dd9e1d2bea0566afecde40641b79
--- /dev/null
+++ b/records/sd1/sweep/4eb2750ea19a1c846014/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4eb2750ea19a1c846014",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:40:27Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.1.transformer_blocks.0.attn1.to_k",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.hi",
+   "kind": "self_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/63139bc30a7ac630912b8cfd620dcd9dced56717e23d6afca4ef0039071951c0.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "63139bc30a7ac630912b8cfd620dcd9dced56717e23d6afca4ef0039071951c0",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "c6ee12d693fd693c651f"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb2750ea19a1c846014/workflow.json b/records/sd1/sweep/4eb2750ea19a1c846014/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..83406e1193990d568fb4f463fb1d813c94db3f9e
--- /dev/null
+++ b/records/sd1/sweep/4eb2750ea19a1c846014/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4eb2750ea19a1c846014)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.1.transformer_blocks.0.attn1.to_k\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb52fa3a311fc7e9f81/measurements.json b/records/sd1/sweep/4eb52fa3a311fc7e9f81/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..8313841dbcdd24b56470c870a615dc64f2dd2668
--- /dev/null
+++ b/records/sd1/sweep/4eb52fa3a311fc7e9f81/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4eb52fa3a311fc7e9f81",
+ "values": {
+  "std": {
+   "value": 55.954,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00978,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 28.874,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.254483,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.282531,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.067543,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.036794,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9635,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2462,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb52fa3a311fc7e9f81/record.json b/records/sd1/sweep/4eb52fa3a311fc7e9f81/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9e41d29489f420e7c84c46d7bab0817042ac2b5c
--- /dev/null
+++ b/records/sd1/sweep/4eb52fa3a311fc7e9f81/record.json
@@ -0,0 +1,77 @@
+{
+ "id": "4eb52fa3a311fc7e9f81",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:40:32Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "grow",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "scale",
+   1.4,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb52fa3a311fc7e9f81/workflow.json b/records/sd1/sweep/4eb52fa3a311fc7e9f81/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..617b1708dca1c5de49a3665b23d70b7d086939a4
--- /dev/null
+++ b/records/sd1/sweep/4eb52fa3a311fc7e9f81/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4eb52fa3a311fc7e9f81)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb530c55ede7a4c8c05/measurements.json b/records/sd1/sweep/4eb530c55ede7a4c8c05/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..13a779beed17dc5810ae0cae9cda35659c7d7267
--- /dev/null
+++ b/records/sd1/sweep/4eb530c55ede7a4c8c05/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4eb530c55ede7a4c8c05",
+ "values": {
+  "std": {
+   "value": 32.359,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02108,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 4.594,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.005866,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.02473,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.012289,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.005143,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0027,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2656,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb530c55ede7a4c8c05/record.json b/records/sd1/sweep/4eb530c55ede7a4c8c05/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b1964d1dae65b468a2376de2fba2326aa270f791
--- /dev/null
+++ b/records/sd1/sweep/4eb530c55ede7a4c8c05/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4eb530c55ede7a4c8c05",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:20:23Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": -0.8
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "mid",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "add_scalar",
+   -0.8,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "fb3ab97f4ee1afbdf792"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb530c55ede7a4c8c05/workflow.json b/records/sd1/sweep/4eb530c55ede7a4c8c05/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..55f84ab5b5093ecc9e3182fb39ba3673089797b9
--- /dev/null
+++ b/records/sd1/sweep/4eb530c55ede7a4c8c05/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4eb530c55ede7a4c8c05)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": -0.8\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb93c3bfffe35414d1f/measurements.json b/records/sd1/sweep/4eb93c3bfffe35414d1f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..553a6bb59eb4c6036e2235ced0d158790ef7bf84
--- /dev/null
+++ b/records/sd1/sweep/4eb93c3bfffe35414d1f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4eb93c3bfffe35414d1f",
+ "values": {
+  "std": {
+   "value": 12.534,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0501,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 32.834,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.512277,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.806208,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.95815,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.449115,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.6104,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.8755,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb93c3bfffe35414d1f/record.json b/records/sd1/sweep/4eb93c3bfffe35414d1f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..065b9d51fadbfb1ab07002176db605d0d4567813
--- /dev/null
+++ b/records/sd1/sweep/4eb93c3bfffe35414d1f/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "4eb93c3bfffe35414d1f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:10:42Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "out.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.mid",
+   "multiply",
+   0.0,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for out.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "0ff553cfafc11b930706"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4eb93c3bfffe35414d1f/workflow.json b/records/sd1/sweep/4eb93c3bfffe35414d1f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c3fd82fb4ea89ee614e363e513d5c8e07876699c
--- /dev/null
+++ b/records/sd1/sweep/4eb93c3bfffe35414d1f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4eb93c3bfffe35414d1f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ee15dca104d749664a8/measurements.json b/records/sd1/sweep/4ee15dca104d749664a8/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..aae110d0a1be60dd2500dca6a11f3024bb2e31de
--- /dev/null
+++ b/records/sd1/sweep/4ee15dca104d749664a8/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4ee15dca104d749664a8",
+ "values": {
+  "cosine_distance": {
+   "value": 0.240565,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.334345,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.206942,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.126973,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 49.381,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00412,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 29.315,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.947,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0141,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ee15dca104d749664a8/record.json b/records/sd1/sweep/4ee15dca104d749664a8/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..44292d1b3ba3a1dc459afa8f139f27ff1a387b87
--- /dev/null
+++ b/records/sd1/sweep/4ee15dca104d749664a8/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4ee15dca104d749664a8",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:32Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.7.1.proj_out",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "in.lo",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/1641e69323523f425ba618385a7095075fdb1a6c7ac31de896a3ac1569191af7.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "1641e69323523f425ba618385a7095075fdb1a6c7ac31de896a3ac1569191af7",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "d5b6569d5a44e0f626bf"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ee15dca104d749664a8/workflow.json b/records/sd1/sweep/4ee15dca104d749664a8/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..90c5309c6b5f40ea978480f9367d5a0430929231
--- /dev/null
+++ b/records/sd1/sweep/4ee15dca104d749664a8/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4ee15dca104d749664a8)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.7.1.proj_out\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ee5a7ee68c8b0575171/measurements.json b/records/sd1/sweep/4ee5a7ee68c8b0575171/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c5c109f01398ed9e647551e4c122f7a8ef11c581
--- /dev/null
+++ b/records/sd1/sweep/4ee5a7ee68c8b0575171/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4ee5a7ee68c8b0575171",
+ "values": {
+  "cosine_distance": {
+   "value": 0.11518,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.140676,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.056635,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.034065,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 56.042,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00571,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 15.692,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9781,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0901,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ee5a7ee68c8b0575171/record.json b/records/sd1/sweep/4ee5a7ee68c8b0575171/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7190edd10fbbb557f55c5103d948a856db086476
--- /dev/null
+++ b/records/sd1/sweep/4ee5a7ee68c8b0575171/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4ee5a7ee68c8b0575171",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:42:45Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.10.0.out_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/7caa77d3991052e5947b9bdbcf067a97747b7935ab97e9b1370fdbd693921727.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "7caa77d3991052e5947b9bdbcf067a97747b7935ab97e9b1370fdbd693921727",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "f533a51027704ac62331"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4ee5a7ee68c8b0575171/workflow.json b/records/sd1/sweep/4ee5a7ee68c8b0575171/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c4df349692d266db13f66d09cbe36289192649c8
--- /dev/null
+++ b/records/sd1/sweep/4ee5a7ee68c8b0575171/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4ee5a7ee68c8b0575171)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.10.0.out_layers.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f2020a834edc82755e2/measurements.json b/records/sd1/sweep/4f2020a834edc82755e2/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..bf9ff0abd993c7ad4fa6e6f4fb360988d6b5d519
--- /dev/null
+++ b/records/sd1/sweep/4f2020a834edc82755e2/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f2020a834edc82755e2",
+ "values": {
+  "cosine_distance": {
+   "value": 0.374787,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.483054,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.330047,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.250021,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 42.172,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00354,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 39.171,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9616,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3784,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f2020a834edc82755e2/record.json b/records/sd1/sweep/4f2020a834edc82755e2/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..e905ec8047c65d5cbc57922361fbf3d0d846f410
--- /dev/null
+++ b/records/sd1/sweep/4f2020a834edc82755e2/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4f2020a834edc82755e2",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.0.0.out_layers.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/4cfa4780e6dcd9b288e01f5b729b28755018ddc816c601e6f4f2870ba9fa9ae4.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "4cfa4780e6dcd9b288e01f5b729b28755018ddc816c601e6f4f2870ba9fa9ae4",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "7c17a0044466693df4a5"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f2020a834edc82755e2/workflow.json b/records/sd1/sweep/4f2020a834edc82755e2/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..4d534c03767b2404b08d7c6b1e586849977b6d6a
--- /dev/null
+++ b/records/sd1/sweep/4f2020a834edc82755e2/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f2020a834edc82755e2)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.0.0.out_layers.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f35c6632b52f94f8431/measurements.json b/records/sd1/sweep/4f35c6632b52f94f8431/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f4bd387dec06bd1999e5f8e1282aab4fc19ba9ae
--- /dev/null
+++ b/records/sd1/sweep/4f35c6632b52f94f8431/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4f35c6632b52f94f8431",
+ "values": {
+  "cosine_distance": {
+   "value": 0.070496,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.096452,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.051231,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.02352,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 63.071,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0051,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 13.444,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0642,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0953,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f35c6632b52f94f8431/record.json b/records/sd1/sweep/4f35c6632b52f94f8431/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7b9eb49aefa29699377837e1119f5b914b58ad97
--- /dev/null
+++ b/records/sd1/sweep/4f35c6632b52f94f8431/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4f35c6632b52f94f8431",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:47:35Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.11.1.transformer_blocks.0.attn2.to_out.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.hi",
+   "kind": "cross_attn",
+   "module_type": "Dropout",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/30d98fcaaa27125b06e52f11b1848b25d554a7c94d2fd379fd1b94fe2e786f07.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "30d98fcaaa27125b06e52f11b1848b25d554a7c94d2fd379fd1b94fe2e786f07",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "963c38583a7da7023413"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f35c6632b52f94f8431/workflow.json b/records/sd1/sweep/4f35c6632b52f94f8431/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..603e05e2dbdf79a3fe284f1008b76136ffc7a789
--- /dev/null
+++ b/records/sd1/sweep/4f35c6632b52f94f8431/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f35c6632b52f94f8431)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.11.1.transformer_blocks.0.attn2.to_out.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f505ad4944c37abe6d5/measurements.json b/records/sd1/sweep/4f505ad4944c37abe6d5/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d4c335d2e27f5bc9d8159c288697280d76498a2
--- /dev/null
+++ b/records/sd1/sweep/4f505ad4944c37abe6d5/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4f505ad4944c37abe6d5",
+ "values": {
+  "cosine_distance": {
+   "value": 0.297422,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.320136,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.230047,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.176731,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.595,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00528,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 37.387,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9549,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.6901,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f505ad4944c37abe6d5/record.json b/records/sd1/sweep/4f505ad4944c37abe6d5/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b97d1e0f101f781894ee088ac98a2deb870da465
--- /dev/null
+++ b/records/sd1/sweep/4f505ad4944c37abe6d5/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4f505ad4944c37abe6d5",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:48:25Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.1.transformer_blocks.0.attn1.to_v",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "in.hi",
+   "kind": "self_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/70d3dddf720901ca81cdde2fff85dd502ed738e40ec7b868bb9cfb448fd5b835.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "70d3dddf720901ca81cdde2fff85dd502ed738e40ec7b868bb9cfb448fd5b835",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "b291c83c8267410af0bb"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f505ad4944c37abe6d5/workflow.json b/records/sd1/sweep/4f505ad4944c37abe6d5/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ce4ac9135af1550768514a170ffc7756456d193
--- /dev/null
+++ b/records/sd1/sweep/4f505ad4944c37abe6d5/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f505ad4944c37abe6d5)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.1.transformer_blocks.0.attn1.to_v\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f53be8fb47725b4dea6/measurements.json b/records/sd1/sweep/4f53be8fb47725b4dea6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..30cd5555912a2cad5a96952a505654c34bba3cd2
--- /dev/null
+++ b/records/sd1/sweep/4f53be8fb47725b4dea6/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f53be8fb47725b4dea6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.294572,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.455707,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.321958,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.207361,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 43.782,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00364,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 32.936,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0035,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2906,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f53be8fb47725b4dea6/record.json b/records/sd1/sweep/4f53be8fb47725b4dea6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..06e405a83df28a15045ee224fa783082d10568e9
--- /dev/null
+++ b/records/sd1/sweep/4f53be8fb47725b4dea6/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4f53be8fb47725b4dea6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:40Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0.skip_connection",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "in.lo",
+   "kind": "skip",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/c2a563faec7a2ec9499883efe28a0613410b37ee94c7826b163b78522ecb3246.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "c2a563faec7a2ec9499883efe28a0613410b37ee94c7826b163b78522ecb3246",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "b214c342fe47a20e4d04"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f53be8fb47725b4dea6/workflow.json b/records/sd1/sweep/4f53be8fb47725b4dea6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..98b6c5d627f70b7b9b913d011cf899821a1a521e
--- /dev/null
+++ b/records/sd1/sweep/4f53be8fb47725b4dea6/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f53be8fb47725b4dea6)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0.skip_connection\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f561af969d8a173c44f/measurements.json b/records/sd1/sweep/4f561af969d8a173c44f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..a00213f1d966a16a80a9962bbe531a482f893110
--- /dev/null
+++ b/records/sd1/sweep/4f561af969d8a173c44f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f561af969d8a173c44f",
+ "values": {
+  "cosine_distance": {
+   "value": 0.132079,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.277955,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.216959,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.11253,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 56.092,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00739,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 18.343,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0086,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0888,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f561af969d8a173c44f/record.json b/records/sd1/sweep/4f561af969d8a173c44f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ddd085cd494a4066629c0e9e42030d8f8c614802
--- /dev/null
+++ b/records/sd1/sweep/4f561af969d8a173c44f/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4f561af969d8a173c44f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:50:35Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.0.in_layers.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/f1eecc0e59dc75992f1a15c3035a1fcf05ca91e069e1cd770cd22c8a3dbefe2d.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "f1eecc0e59dc75992f1a15c3035a1fcf05ca91e069e1cd770cd22c8a3dbefe2d",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "677f549cc104fc26e825"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f561af969d8a173c44f/workflow.json b/records/sd1/sweep/4f561af969d8a173c44f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..4effbbbef5b706d98329529f5141354273d9b180
--- /dev/null
+++ b/records/sd1/sweep/4f561af969d8a173c44f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f561af969d8a173c44f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.0.in_layers.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f673d24f05d7cfa87ad/measurements.json b/records/sd1/sweep/4f673d24f05d7cfa87ad/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..4ed659dc0de5808d8b3799ce1e740b8eefff87a8
--- /dev/null
+++ b/records/sd1/sweep/4f673d24f05d7cfa87ad/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f673d24f05d7cfa87ad",
+ "values": {
+  "cosine_distance": {
+   "value": 0.277185,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.311472,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.127887,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.045181,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 47.812,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00543,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 31.427,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9426,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1124,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f673d24f05d7cfa87ad/record.json b/records/sd1/sweep/4f673d24f05d7cfa87ad/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5548351e064623503c08c0642d1327422c4b05dd
--- /dev/null
+++ b/records/sd1/sweep/4f673d24f05d7cfa87ad/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4f673d24f05d7cfa87ad",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:41:22Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0.in_layers.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/f05a55d1478ab91e990ddb9aafdb24fa4aefb1270a22a7362972b5a3e1a0657f.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "f05a55d1478ab91e990ddb9aafdb24fa4aefb1270a22a7362972b5a3e1a0657f",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "1e67d499721e41408c97"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f673d24f05d7cfa87ad/workflow.json b/records/sd1/sweep/4f673d24f05d7cfa87ad/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ae5ed2af47bc137ae828171af8dadd93df3420b8
--- /dev/null
+++ b/records/sd1/sweep/4f673d24f05d7cfa87ad/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f673d24f05d7cfa87ad)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0.in_layers.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f676d697bb3cb2d4901/measurements.json b/records/sd1/sweep/4f676d697bb3cb2d4901/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..3fde97d991ce4fd895e1b394519674579ec42886
--- /dev/null
+++ b/records/sd1/sweep/4f676d697bb3cb2d4901/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f676d697bb3cb2d4901",
+ "values": {
+  "cosine_distance": {
+   "value": 0.414331,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.417312,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.238866,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.096768,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.139,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00785,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 40.391,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0097,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2548,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f676d697bb3cb2d4901/record.json b/records/sd1/sweep/4f676d697bb3cb2d4901/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d67bd7f075c7d52659baf03a1cf7e50484c58649
--- /dev/null
+++ b/records/sd1/sweep/4f676d697bb3cb2d4901/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4f676d697bb3cb2d4901",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:00Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.11.0.out_layers.1",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/a70053592d0b9e324b82caf387d3712b5723540ebb14e43f2d84b955c1c5446c.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "a70053592d0b9e324b82caf387d3712b5723540ebb14e43f2d84b955c1c5446c",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "85c1a5d4df1f4e41d021"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f676d697bb3cb2d4901/workflow.json b/records/sd1/sweep/4f676d697bb3cb2d4901/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ba20d7cffaebd3ce70585ef45984ed577cca60e8
--- /dev/null
+++ b/records/sd1/sweep/4f676d697bb3cb2d4901/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f676d697bb3cb2d4901)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.11.0.out_layers.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f7c046ea063a84a4fc7/measurements.json b/records/sd1/sweep/4f7c046ea063a84a4fc7/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..3d88d4c6a3a16c71b96f33a685558bed47aa0db9
--- /dev/null
+++ b/records/sd1/sweep/4f7c046ea063a84a4fc7/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f7c046ea063a84a4fc7",
+ "values": {
+  "cosine_distance": {
+   "value": 0.031582,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.052664,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.017879,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.010879,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.444,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00636,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 7.697,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0108,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1634,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f7c046ea063a84a4fc7/record.json b/records/sd1/sweep/4f7c046ea063a84a4fc7/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..e949160555b6df71490df71e07452b260483c33c
--- /dev/null
+++ b/records/sd1/sweep/4f7c046ea063a84a4fc7/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4f7c046ea063a84a4fc7",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:07Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.1.transformer_blocks.0.ff.net.2",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "out.mid",
+   "kind": "ff",
+   "module_type": "Linear",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/40e420d6a2363675b28efa58d2d180ebe58e793cd7b40a1bc714ad78b3a592f1.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "40e420d6a2363675b28efa58d2d180ebe58e793cd7b40a1bc714ad78b3a592f1",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.FeedForward.Sequential.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e74b46fb0ace639910fb"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f7c046ea063a84a4fc7/workflow.json b/records/sd1/sweep/4f7c046ea063a84a4fc7/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..734dc0c1bfb6b3310ce96f7284fcb978e3914c93
--- /dev/null
+++ b/records/sd1/sweep/4f7c046ea063a84a4fc7/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f7c046ea063a84a4fc7)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.1.transformer_blocks.0.ff.net.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f873fee659c31f01625/measurements.json b/records/sd1/sweep/4f873fee659c31f01625/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..3d87cc8306c53f9b08bf3c78fe30229ad0ac0ef4
--- /dev/null
+++ b/records/sd1/sweep/4f873fee659c31f01625/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4f873fee659c31f01625",
+ "values": {
+  "cosine_distance": {
+   "value": 0.0,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.0,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.618,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00534,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.0,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1553,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f873fee659c31f01625/record.json b/records/sd1/sweep/4f873fee659c31f01625/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..c3ffef12d380cd4fb06a94ca60c431afb4340c0a
--- /dev/null
+++ b/records/sd1/sweep/4f873fee659c31f01625/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4f873fee659c31f01625",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:42:20Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0.h_upd",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/7b180bf44edea53419df643591dd183ce23a65e67a2472d3dda41047249b306d.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "7b180bf44edea53419df643591dd183ce23a65e67a2472d3dda41047249b306d",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "ac5c8f0b5a8b1b63480c"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f873fee659c31f01625/workflow.json b/records/sd1/sweep/4f873fee659c31f01625/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c9a78f908d5e8989c58f978005db24911f1680c1
--- /dev/null
+++ b/records/sd1/sweep/4f873fee659c31f01625/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f873fee659c31f01625)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0.h_upd\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f9281bc42857aea6ec2/measurements.json b/records/sd1/sweep/4f9281bc42857aea6ec2/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e23d88a43a5ec3cf9750c271d0d50f9f5d8c75a9
--- /dev/null
+++ b/records/sd1/sweep/4f9281bc42857aea6ec2/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f9281bc42857aea6ec2",
+ "values": {
+  "std": {
+   "value": 31.592,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02481,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 28.853,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.28682,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.477146,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.175454,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.048178,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9527,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.288,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f9281bc42857aea6ec2/record.json b/records/sd1/sweep/4f9281bc42857aea6ec2/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a7400cf285dda74b3fa99b638d4bb38e9c9396b8
--- /dev/null
+++ b/records/sd1/sweep/4f9281bc42857aea6ec2/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "4f9281bc42857aea6ec2",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:57:59Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180.0
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "half",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "rotate",
+   180.0,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "1bf3d1dd008445d22415"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f9281bc42857aea6ec2/workflow.json b/records/sd1/sweep/4f9281bc42857aea6ec2/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a4ed30bfd1dd5527b62f44f4b13f4015b47c0539
--- /dev/null
+++ b/records/sd1/sweep/4f9281bc42857aea6ec2/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f9281bc42857aea6ec2)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f931808ca93138e3d55/interpretations.jsonl b/records/sd1/sweep/4f931808ca93138e3d55/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..af3a6806f293af3ed028e4fe653f4dbb8d6bab76
--- /dev/null
+++ b/records/sd1/sweep/4f931808ca93138e3d55/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "4f931808ca93138e3d55"}, "kind": "caption", "value": "Dense speckled pattern of dark, white and teal flecks", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:51:01Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "1b4071a38d80358a"}
+{"target": {"record": "4f931808ca93138e3d55"}, "kind": "change", "value": "The portrait vanishes into a busy, granite-like speckle of dark and light fragments with hints of blue and green.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:51:01Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "999c143ddf0bc890"}
+{"target": {"record": "4f931808ca93138e3d55"}, "kind": "keywords", "value": ["speckled texture", "granite pattern", "lost portrait", "abstract flecks"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:51:01Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "456f5f9d8a1e5470"}
+{"target": {"record": "4f931808ca93138e3d55"}, "kind": "effect_tags", "value": ["abstract", "subject_lost", "texture", "pattern", "degenerate"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:51:01Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "d8579e2d256b0cb4"}
diff --git a/records/sd1/sweep/4f931808ca93138e3d55/measurements.json b/records/sd1/sweep/4f931808ca93138e3d55/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..9e70be6ac895908e4fd102bdf052fce48ce99a21
--- /dev/null
+++ b/records/sd1/sweep/4f931808ca93138e3d55/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f931808ca93138e3d55",
+ "values": {
+  "std": {
+   "value": 67.476,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0089,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 63.225,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.600589,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.65538,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.961939,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.475985,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.6271,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1762,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f931808ca93138e3d55/record.json b/records/sd1/sweep/4f931808ca93138e3d55/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..54a53ae169446f39dcba7c0808475b8a3e4f2d85
--- /dev/null
+++ b/records/sd1/sweep/4f931808ca93138e3d55/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "4f931808ca93138e3d55",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:31:20Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180.0
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "half",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "2976534844fe80be26b4",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "rotate",
+   180.0,
+   "middle"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 180.0,
+    "applied": 180.0,
+    "reason": "outside safe range [0.0, 5.0] for out.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "032862a33ab44957d380"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f931808ca93138e3d55/workflow.json b/records/sd1/sweep/4f931808ca93138e3d55/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..773b799b52bbc6d2f179344a764d0eb064027aba
--- /dev/null
+++ b/records/sd1/sweep/4f931808ca93138e3d55/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f931808ca93138e3d55)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f938d32dad76a588083/measurements.json b/records/sd1/sweep/4f938d32dad76a588083/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c1db2381cee425575b1456e7e04c26c927a49bb
--- /dev/null
+++ b/records/sd1/sweep/4f938d32dad76a588083/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4f938d32dad76a588083",
+ "values": {
+  "std": {
+   "value": 36.775,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00836,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 41.868,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.444635,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.667958,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.652512,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.253899,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.7896,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0351,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f938d32dad76a588083/record.json b/records/sd1/sweep/4f938d32dad76a588083/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..2afb70d5b67a0fdecc3fcf67ce0f1e51dbebb57a
--- /dev/null
+++ b/records/sd1/sweep/4f938d32dad76a588083/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "4f938d32dad76a588083",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:03:22Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "13216623a742ccb32c5e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "rotate",
+   180.0,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 180.0,
+    "applied": 180.0,
+    "reason": "outside safe range [5.0, 45.0] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "463833e1942b543c7275"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4f938d32dad76a588083/workflow.json b/records/sd1/sweep/4f938d32dad76a588083/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..23391e46a5f46932e1cc6f26cacf3f2e20a2e6db
--- /dev/null
+++ b/records/sd1/sweep/4f938d32dad76a588083/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4f938d32dad76a588083)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180.0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fb01994418997711455/measurements.json b/records/sd1/sweep/4fb01994418997711455/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e664f9aee2136c7ed281ecc278b0d9169bcbabf2
--- /dev/null
+++ b/records/sd1/sweep/4fb01994418997711455/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4fb01994418997711455",
+ "values": {
+  "cosine_distance": {
+   "value": 0.044912,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.075535,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.02341,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.016366,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.329,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0062,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 10.569,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9893,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1844,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fb01994418997711455/record.json b/records/sd1/sweep/4fb01994418997711455/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1112108057a4f75fe30cc4e6be8920a62c8eb646
--- /dev/null
+++ b/records/sd1/sweep/4fb01994418997711455/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4fb01994418997711455",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:43:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.0.0.out_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 1.5
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/6998d497e0ce384bf7ca04cd0824ca7047ae7f5533ed77446fd139d529d2622a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "6998d497e0ce384bf7ca04cd0824ca7047ae7f5533ed77446fd139d529d2622a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "83e9df7f95f3a3a8f807"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fb01994418997711455/workflow.json b/records/sd1/sweep/4fb01994418997711455/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..dd1c5744625b614660ae2c7497eeaafa0863b274
--- /dev/null
+++ b/records/sd1/sweep/4fb01994418997711455/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fb01994418997711455)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.0.0.out_layers.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fbd57bfde9c74b0dc4e/measurements.json b/records/sd1/sweep/4fbd57bfde9c74b0dc4e/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6cd29a9c9d14abc28d3442d5615411cfbd6b399b
--- /dev/null
+++ b/records/sd1/sweep/4fbd57bfde9c74b0dc4e/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4fbd57bfde9c74b0dc4e",
+ "values": {
+  "cosine_distance": {
+   "value": 0.041919,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.070818,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.026828,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.00908,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.22,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00543,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 8.882,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9972,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1546,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fbd57bfde9c74b0dc4e/record.json b/records/sd1/sweep/4fbd57bfde9c74b0dc4e/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d7b92fb775de627bf9c03b0de53a0b99ca0bb6b
--- /dev/null
+++ b/records/sd1/sweep/4fbd57bfde9c74b0dc4e/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4fbd57bfde9c74b0dc4e",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:43:25Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "middle_block.2.in_layers.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/1c2d564c6d8ccd9f66df3b5b76fff6054130c1cd137e848caa1a481bdd9179fd.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "1c2d564c6d8ccd9f66df3b5b76fff6054130c1cd137e848caa1a481bdd9179fd",
+  "type_path": "UNetModel.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "57421ffc4d6c282d1283"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fbd57bfde9c74b0dc4e/workflow.json b/records/sd1/sweep/4fbd57bfde9c74b0dc4e/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..32ad25224305b4f85a56276ed42fcce888925073
--- /dev/null
+++ b/records/sd1/sweep/4fbd57bfde9c74b0dc4e/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fbd57bfde9c74b0dc4e)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.2.in_layers.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/measurements.json b/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..5675338c741794c9fe76c2c7698a48dd19373204
--- /dev/null
+++ b/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4fbf6a9fbd4d127fd7c6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.160923,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.227389,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.075827,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.033817,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.196,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00522,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 19.426,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9414,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2309,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/record.json b/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..23a8013de56e5d3d0b233354aa978c5de2cfff94
--- /dev/null
+++ b/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4fbf6a9fbd4d127fd7c6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:55Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.emb_layers.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "Linear",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/6512e5d39b6947bcc884712d2e03454026220bcbfd9420a11d58f011a1c495b9.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "6512e5d39b6947bcc884712d2e03454026220bcbfd9420a11d58f011a1c495b9",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "d473cae4bc4ab95fc433"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/workflow.json b/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..590acffba6e6b455fbbfeab800468135d621cda1
--- /dev/null
+++ b/records/sd1/sweep/4fbf6a9fbd4d127fd7c6/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fbf6a9fbd4d127fd7c6)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.emb_layers.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fc6110b0f8a0c9529d0/measurements.json b/records/sd1/sweep/4fc6110b0f8a0c9529d0/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..916618eacd1569cccad62f608ba6d877013ab05b
--- /dev/null
+++ b/records/sd1/sweep/4fc6110b0f8a0c9529d0/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4fc6110b0f8a0c9529d0",
+ "values": {
+  "std": {
+   "value": 32.402,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02137,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 48.442,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.595005,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.554786,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.447803,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.119433,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9954,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0273,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fc6110b0f8a0c9529d0/record.json b/records/sd1/sweep/4fc6110b0f8a0c9529d0/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..2661da36dbe7c0b34810db037c0fb1d46fcd785c
--- /dev/null
+++ b/records/sd1/sweep/4fc6110b0f8a0c9529d0/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "4fc6110b0f8a0c9529d0",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:13:11Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "out.lo",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "zero",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "13216623a742ccb32c5e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.lo",
+   "multiply",
+   0.0,
+   "early"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for out.lo (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "2c86a33c688f2ced80d1"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fc6110b0f8a0c9529d0/workflow.json b/records/sd1/sweep/4fc6110b0f8a0c9529d0/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..aa52b4525c42122e492b8ee4bcad848dda8059eb
--- /dev/null
+++ b/records/sd1/sweep/4fc6110b0f8a0c9529d0/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fc6110b0f8a0c9529d0)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fc9402d44b7a9649d12/interpretations.jsonl b/records/sd1/sweep/4fc9402d44b7a9649d12/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..8e6c426728f68a7e9871acab58009fba215204dd
--- /dev/null
+++ b/records/sd1/sweep/4fc9402d44b7a9649d12/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "4fc9402d44b7a9649d12"}, "kind": "caption", "value": "Illustrated young woman with slicked dark hair on a bright teal gridded background, poster-like.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:27:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "d075ded61f049f95"}
+{"target": {"record": "4fc9402d44b7a9649d12"}, "kind": "change", "value": "The muted photo becomes a saturated illustrated portrait in a different style: a vivid teal grid-textured background, smoothed stylised face and a front-facing pose with sharp, graphic detail.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:27:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "2be4dbd971b855a2"}
+{"target": {"record": "4fc9402d44b7a9649d12"}, "kind": "keywords", "value": ["teal background", "illustrated portrait", "grid texture", "poster style", "stylised face"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:27:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "5c5e7f1fcae3fb63"}
+{"target": {"record": "4fc9402d44b7a9649d12"}, "kind": "effect_tags", "value": ["style_change", "saturated", "colour_shift", "poster", "pattern", "texture", "subject_kept", "bright"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:27:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "1f5fb461a8815cf2"}
diff --git a/records/sd1/sweep/4fc9402d44b7a9649d12/measurements.json b/records/sd1/sweep/4fc9402d44b7a9649d12/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5c3db7ee9183b7451ff9758aeb4416f108f050d
--- /dev/null
+++ b/records/sd1/sweep/4fc9402d44b7a9649d12/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4fc9402d44b7a9649d12",
+ "values": {
+  "cosine_distance": {
+   "value": 0.399222,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.592448,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.346326,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.191743,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 67.274,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.04162,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 59.916,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9161,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.609,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fc9402d44b7a9649d12/record.json b/records/sd1/sweep/4fc9402d44b7a9649d12/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5831b06ba5174ba3cd379f3a7b1bef850b827aee
--- /dev/null
+++ b/records/sd1/sweep/4fc9402d44b7a9649d12/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4fc9402d44b7a9649d12",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:59Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.11.0.emb_layers.1",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "Linear",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/4d50c2c02ae86f255513b08a69eea2b0fa0377c2e55fb3560c9aed3334fdf4f9.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "4d50c2c02ae86f255513b08a69eea2b0fa0377c2e55fb3560c9aed3334fdf4f9",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "befdbddffa8071da56a0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fc9402d44b7a9649d12/workflow.json b/records/sd1/sweep/4fc9402d44b7a9649d12/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d4a9e53aa95a1b8a7799e4cd3f22de7d22371f9
--- /dev/null
+++ b/records/sd1/sweep/4fc9402d44b7a9649d12/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fc9402d44b7a9649d12)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.11.0.emb_layers.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fcea844f68b0673f27c/measurements.json b/records/sd1/sweep/4fcea844f68b0673f27c/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..386af0943a611412f2b9afd1d8f457bf7d548613
--- /dev/null
+++ b/records/sd1/sweep/4fcea844f68b0673f27c/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4fcea844f68b0673f27c",
+ "values": {
+  "cosine_distance": {
+   "value": 0.535617,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.652898,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.673301,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.286328,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 28.966,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00264,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 48.201,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9984,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.211,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fcea844f68b0673f27c/record.json b/records/sd1/sweep/4fcea844f68b0673f27c/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..82faab4397e081bfd45a594a64f12efd65df7f9f
--- /dev/null
+++ b/records/sd1/sweep/4fcea844f68b0673f27c/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4fcea844f68b0673f27c",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:49:41Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.0.out_layers.3",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/fbf73353b9cac4bd0d7c97646a2fb384ad52dda1d35d74bd800103563d510d3a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "fbf73353b9cac4bd0d7c97646a2fb384ad52dda1d35d74bd800103563d510d3a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "deb05fc5262001266be0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fcea844f68b0673f27c/workflow.json b/records/sd1/sweep/4fcea844f68b0673f27c/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..2043e0e865003c8be3fde98a925fcd4d22e6617f
--- /dev/null
+++ b/records/sd1/sweep/4fcea844f68b0673f27c/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fcea844f68b0673f27c)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.0.out_layers.3\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fd50458824b337d7f5e/measurements.json b/records/sd1/sweep/4fd50458824b337d7f5e/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d4dfab26a08ed8de8cbbb11cf93125f75ac07a2
--- /dev/null
+++ b/records/sd1/sweep/4fd50458824b337d7f5e/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4fd50458824b337d7f5e",
+ "values": {
+  "std": {
+   "value": 40.828,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02195,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 34.345,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.303268,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.536266,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.459753,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.186746,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9417,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0261,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fd50458824b337d7f5e/record.json b/records/sd1/sweep/4fd50458824b337d7f5e/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7b6bf2a04b4ffe5f21824df9fd296fc6d851b906
--- /dev/null
+++ b/records/sd1/sweep/4fd50458824b337d7f5e/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "4fd50458824b337d7f5e",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:57:51Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "out.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.mid",
+   "scale",
+   1.4,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 1.4,
+    "applied": 1.4,
+    "reason": "outside safe range [0.7, 1.3] for out.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "08853770c97bfe43c630"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fd50458824b337d7f5e/workflow.json b/records/sd1/sweep/4fd50458824b337d7f5e/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..32c6a69e2c12ad07c029536c920cebfbb95f2bd1
--- /dev/null
+++ b/records/sd1/sweep/4fd50458824b337d7f5e/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fd50458824b337d7f5e)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/measurements.json b/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..d30afd7bbd3ffa787d961f0492e59abeee8d7e62
--- /dev/null
+++ b/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4fdd3d7d8f8384b1b8f2",
+ "values": {
+  "cosine_distance": {
+   "value": 0.532459,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.697545,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.888376,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.503568,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 83.653,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01058,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 66.883,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9007,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.8605,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/record.json b/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a5c0c53d846a8619ab53fdb2d3bccdd4e03c021b
--- /dev/null
+++ b/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4fdd3d7d8f8384b1b8f2",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:32Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.0.skip_connection",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.5
+   },
+   "group": "in.hi",
+   "kind": "skip",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/1cf87c03fcd0683a3b0994efefe1018445a99ae4c2275686bd27dc67b9a7468a.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "1cf87c03fcd0683a3b0994efefe1018445a99ae4c2275686bd27dc67b9a7468a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "186611bc108d6677637b"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/workflow.json b/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..0ed350d468139be80ac5e3afbe4c503146563f5a
--- /dev/null
+++ b/records/sd1/sweep/4fdd3d7d8f8384b1b8f2/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fdd3d7d8f8384b1b8f2)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.0.skip_connection\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fede74f6f474fcb99a8/measurements.json b/records/sd1/sweep/4fede74f6f474fcb99a8/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0f9e61020e48a5cb9ec2c02c5d1719d444179e80
--- /dev/null
+++ b/records/sd1/sweep/4fede74f6f474fcb99a8/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "4fede74f6f474fcb99a8",
+ "values": {
+  "cosine_distance": {
+   "value": 0.407705,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.369619,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.344707,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.256032,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 52.248,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00527,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 42.441,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.975,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5474,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fede74f6f474fcb99a8/record.json b/records/sd1/sweep/4fede74f6f474fcb99a8/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..76e3bc4577fae307848afc095e8699ed041378c6
--- /dev/null
+++ b/records/sd1/sweep/4fede74f6f474fcb99a8/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "4fede74f6f474fcb99a8",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:48:37Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.4.1.transformer_blocks.0.attn1.to_k",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "in.mid",
+   "kind": "self_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/dfd160f73f1ec0c872ab4da9dc300bfe1ef8ed0be85f50fbdf5b943e7704a6f5.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "dfd160f73f1ec0c872ab4da9dc300bfe1ef8ed0be85f50fbdf5b943e7704a6f5",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "4cd7f052e35eff6dba53"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fede74f6f474fcb99a8/workflow.json b/records/sd1/sweep/4fede74f6f474fcb99a8/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f3f5ef69e54bf0d9985fff615f101548b09adf64
--- /dev/null
+++ b/records/sd1/sweep/4fede74f6f474fcb99a8/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fede74f6f474fcb99a8)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.4.1.transformer_blocks.0.attn1.to_k\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fee6f6d9afc3d08f0a6/interpretations.jsonl b/records/sd1/sweep/4fee6f6d9afc3d08f0a6/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..1df5d6a62b53ea2bef2376175d29940f7bd20ded
--- /dev/null
+++ b/records/sd1/sweep/4fee6f6d9afc3d08f0a6/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "4fee6f6d9afc3d08f0a6"}, "kind": "caption", "value": "Pale, soft painted portrait of a woman with light brown hair on grey", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:21:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "16321dcefc38cee0"}
+{"target": {"record": "4fee6f6d9afc3d08f0a6"}, "kind": "change", "value": "The image fades and softens, with lighter ash-brown hair, a gentler face and a pale grey background, keeping the same composition.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:21:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "14ec800ab98c03ff"}
+{"target": {"record": "4fee6f6d9afc3d08f0a6"}, "kind": "keywords", "value": ["faded portrait", "pale tones", "soft face", "ash brown hair"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:21:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "befb05a0a08342f9"}
+{"target": {"record": "4fee6f6d9afc3d08f0a6"}, "kind": "effect_tags", "value": ["subject_kept", "washed_out", "painterly", "subtle"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:21:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "e1a9aaabb350f956"}
diff --git a/records/sd1/sweep/4fee6f6d9afc3d08f0a6/measurements.json b/records/sd1/sweep/4fee6f6d9afc3d08f0a6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..abb6e7773c8655fc85c9d923c6c96fa9d7c458eb
--- /dev/null
+++ b/records/sd1/sweep/4fee6f6d9afc3d08f0a6/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "4fee6f6d9afc3d08f0a6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.178841,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.247265,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.076556,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.055663,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 49.832,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00465,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.726,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9776,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0902,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fee6f6d9afc3d08f0a6/record.json b/records/sd1/sweep/4fee6f6d9afc3d08f0a6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..992c51d44709eb64117312558691830f8d01f030
--- /dev/null
+++ b/records/sd1/sweep/4fee6f6d9afc3d08f0a6/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "4fee6f6d9afc3d08f0a6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:18Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "in.mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/467a46968dde45b4234dd158ebfdbd4a3da9827686daaca612af01440b160e09.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "467a46968dde45b4234dd158ebfdbd4a3da9827686daaca612af01440b160e09",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "bfe722f3f051092602bb"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/4fee6f6d9afc3d08f0a6/workflow.json b/records/sd1/sweep/4fee6f6d9afc3d08f0a6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f17816f8e6ab2e9f61737db4bdd165e5dde717c2
--- /dev/null
+++ b/records/sd1/sweep/4fee6f6d9afc3d08f0a6/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 4fee6f6d9afc3d08f0a6)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/500f05aa8533434ce81a/measurements.json b/records/sd1/sweep/500f05aa8533434ce81a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..1a4b49f3509e363759976796891e1cd721ec19e8
--- /dev/null
+++ b/records/sd1/sweep/500f05aa8533434ce81a/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "500f05aa8533434ce81a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.338455,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.374898,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.149025,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.071417,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.149,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00277,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 39.097,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9386,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2062,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/500f05aa8533434ce81a/record.json b/records/sd1/sweep/500f05aa8533434ce81a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9ea68d00ebb0bedb49f321d96667d01ccfcdadc7
--- /dev/null
+++ b/records/sd1/sweep/500f05aa8533434ce81a/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "500f05aa8533434ce81a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:29Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.7.0.out_layers.3",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/0914605d8f7db9a4bde9bfcaaee6ec5de8be9d295e104ba14daae626f49353af.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "0914605d8f7db9a4bde9bfcaaee6ec5de8be9d295e104ba14daae626f49353af",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "4a539e6cdd58d97433a2"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/500f05aa8533434ce81a/workflow.json b/records/sd1/sweep/500f05aa8533434ce81a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c74cca0c8f51272c3b52637fae41605cc1917e55
--- /dev/null
+++ b/records/sd1/sweep/500f05aa8533434ce81a/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 500f05aa8533434ce81a)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.7.0.out_layers.3\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50122ec67794d649a5a6/measurements.json b/records/sd1/sweep/50122ec67794d649a5a6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5b6b201432e267bc451848ca06048b2d48a18c0
--- /dev/null
+++ b/records/sd1/sweep/50122ec67794d649a5a6/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50122ec67794d649a5a6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.0,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.0,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": -0.0,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 66.477,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00637,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.0,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0093,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2957,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50122ec67794d649a5a6/record.json b/records/sd1/sweep/50122ec67794d649a5a6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..551aa92421b586a8635dac5f9b18a03f8a949c5b
--- /dev/null
+++ b/records/sd1/sweep/50122ec67794d649a5a6/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "50122ec67794d649a5a6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:50Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.2.0.h_upd",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/b2eee590eb995120f187358683de01f17fde9dc5d694b2983b76f48b05fb695d.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "b2eee590eb995120f187358683de01f17fde9dc5d694b2983b76f48b05fb695d",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "fb32b39e919e38b42483"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50122ec67794d649a5a6/workflow.json b/records/sd1/sweep/50122ec67794d649a5a6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f8a4c12bfd963e75c9f91ca8a1e00e02350fc9a9
--- /dev/null
+++ b/records/sd1/sweep/50122ec67794d649a5a6/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50122ec67794d649a5a6)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.2.0.h_upd\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5015df82df1ef5fd4bfb/measurements.json b/records/sd1/sweep/5015df82df1ef5fd4bfb/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cb7eb82f404a226fe7d2cbbd4dda8a2582c42e20
--- /dev/null
+++ b/records/sd1/sweep/5015df82df1ef5fd4bfb/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5015df82df1ef5fd4bfb",
+ "values": {
+  "std": {
+   "value": 28.564,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02515,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 33.562,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.422205,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.561608,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.53676,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.165575,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.8247,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.022,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5015df82df1ef5fd4bfb/record.json b/records/sd1/sweep/5015df82df1ef5fd4bfb/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1469bcd9d91e70adc390fdd7a770ad6e386202c3
--- /dev/null
+++ b/records/sd1/sweep/5015df82df1ef5fd4bfb/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "5015df82df1ef5fd4bfb",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:03:32Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.7
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "shrink",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "scale",
+   0.7,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "d7b212e3c6f0fc7fb75d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5015df82df1ef5fd4bfb/workflow.json b/records/sd1/sweep/5015df82df1ef5fd4bfb/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..8e3047f8dec477217112f3aac554ad0029e4b400
--- /dev/null
+++ b/records/sd1/sweep/5015df82df1ef5fd4bfb/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5015df82df1ef5fd4bfb)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.7\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/501882c96275897a4e2f/measurements.json b/records/sd1/sweep/501882c96275897a4e2f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..15a42030e64a5b9df151664ad0819eef034a749c
--- /dev/null
+++ b/records/sd1/sweep/501882c96275897a4e2f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "501882c96275897a4e2f",
+ "values": {
+  "std": {
+   "value": 50.979,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0109,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 4.649,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.014509,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.065098,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.061054,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0084,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.999,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0085,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/501882c96275897a4e2f/record.json b/records/sd1/sweep/501882c96275897a4e2f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c6d9d360b4b40f179d80d52d13a6df931dd90d0
--- /dev/null
+++ b/records/sd1/sweep/501882c96275897a4e2f/record.json
@@ -0,0 +1,79 @@
+{
+ "id": "501882c96275897a4e2f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:35:20Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.8,
+    "seed": 1234
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "heavy",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "add_noise",
+   0.8,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "858fa7df246e04b0367c"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/501882c96275897a4e2f/workflow.json b/records/sd1/sweep/501882c96275897a4e2f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..d85867a2abb3dec3b82a108f6e722667c05b06b7
--- /dev/null
+++ b/records/sd1/sweep/501882c96275897a4e2f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 501882c96275897a4e2f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.8,\n    \"seed\": 1234\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50238b0458a13f63776d/measurements.json b/records/sd1/sweep/50238b0458a13f63776d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..1bb14100c9dd45faf0cd07f0fb287611109f6588
--- /dev/null
+++ b/records/sd1/sweep/50238b0458a13f63776d/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50238b0458a13f63776d",
+ "values": {
+  "std": {
+   "value": 37.732,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00595,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 40.539,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.378084,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.692116,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.877288,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.328443,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.6166,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.8443,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50238b0458a13f63776d/record.json b/records/sd1/sweep/50238b0458a13f63776d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..8616028d8cf5bd45e68062cb49399cf5b92e63a1
--- /dev/null
+++ b/records/sd1/sweep/50238b0458a13f63776d/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "50238b0458a13f63776d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:50:51Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30.0
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "slight",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "rotate",
+   30.0,
+   "middle"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 30.0,
+    "applied": 30.0,
+    "reason": "outside safe range [0.0, 5.0] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "68161aeb49eeae0d33ef"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50238b0458a13f63776d/workflow.json b/records/sd1/sweep/50238b0458a13f63776d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a942ef0d535a34f887530d81530c2eebec713880
--- /dev/null
+++ b/records/sd1/sweep/50238b0458a13f63776d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50238b0458a13f63776d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50294fc4f301315f400d/interpretations.jsonl b/records/sd1/sweep/50294fc4f301315f400d/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..a673fc41330299c5e6e65d1b4b98235293bf58ab
--- /dev/null
+++ b/records/sd1/sweep/50294fc4f301315f400d/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "50294fc4f301315f400d"}, "kind": "caption", "value": "Close-up of a wrinkled old face with closed eyes and a long grey beard", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:57:53Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3b22590ed3e2774a"}
+{"target": {"record": "50294fc4f301315f400d"}, "kind": "change", "value": "The view moves closer to the face and the eyes appear closed or squinting, with heavier wrinkles and a smudged, ghostly overlay of texture.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:57:53Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "f56d9be85649b7f3"}
+{"target": {"record": "50294fc4f301315f400d"}, "kind": "keywords", "value": ["close-up face", "closed eyes", "wrinkled skin", "grey beard", "ghostly overlay"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:57:53Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a1b7462ef7b16e02"}
+{"target": {"record": "50294fc4f301315f400d"}, "kind": "effect_tags", "value": ["zoom", "subject_kept", "texture", "distorted"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:57:53Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "24f22d784ef6f45f"}
diff --git a/records/sd1/sweep/50294fc4f301315f400d/measurements.json b/records/sd1/sweep/50294fc4f301315f400d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..18db720bd3dd3da10ed223ed86054ff288d468ae
--- /dev/null
+++ b/records/sd1/sweep/50294fc4f301315f400d/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50294fc4f301315f400d",
+ "values": {
+  "std": {
+   "value": 44.714,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02093,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 34.737,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.403824,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.521218,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.217236,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.138997,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.899,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.6225,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50294fc4f301315f400d/record.json b/records/sd1/sweep/50294fc4f301315f400d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..469de197b0a9736167943fa6021f715e8b3d1a4b
--- /dev/null
+++ b/records/sd1/sweep/50294fc4f301315f400d/record.json
@@ -0,0 +1,84 @@
+{
+ "id": "50294fc4f301315f400d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:20:05Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "grow",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "2976534844fe80be26b4",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "scale",
+   1.4,
+   "middle"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 1.4,
+    "applied": 1.4,
+    "reason": "outside safe range [0.7, 1.3] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50294fc4f301315f400d/workflow.json b/records/sd1/sweep/50294fc4f301315f400d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..12d7327a1f474ad254a6cdfe15dc3b2afd0e0611
--- /dev/null
+++ b/records/sd1/sweep/50294fc4f301315f400d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50294fc4f301315f400d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/502b666eaba4001d1478/measurements.json b/records/sd1/sweep/502b666eaba4001d1478/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..8bd4a251278c6b4d0885d484705fd9b5fb284d67
--- /dev/null
+++ b/records/sd1/sweep/502b666eaba4001d1478/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "502b666eaba4001d1478",
+ "values": {
+  "cosine_distance": {
+   "value": 0.245259,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.314608,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.280931,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.177216,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.52,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00299,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 27.057,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0257,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0801,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/502b666eaba4001d1478/record.json b/records/sd1/sweep/502b666eaba4001d1478/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7977a559cb3b4e08d002439da848b95df23be107
--- /dev/null
+++ b/records/sd1/sweep/502b666eaba4001d1478/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "502b666eaba4001d1478",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:13Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.5.0.out_layers.3",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.5
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/dd6c25afbf76002d37835911dcbd61788666dded13ba3b501f44749eeace32f6.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "dd6c25afbf76002d37835911dcbd61788666dded13ba3b501f44749eeace32f6",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "dfba29a0bef8a40cda0e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/502b666eaba4001d1478/workflow.json b/records/sd1/sweep/502b666eaba4001d1478/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..455d9827302e3a3cc1c81e933215da8c8b1f4dca
--- /dev/null
+++ b/records/sd1/sweep/502b666eaba4001d1478/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 502b666eaba4001d1478)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.5.0.out_layers.3\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/502caea572f671437152/interpretations.jsonl b/records/sd1/sweep/502caea572f671437152/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..5e6e03bb010ecec950c57d35b225a15859f8c662
--- /dev/null
+++ b/records/sd1/sweep/502caea572f671437152/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "502caea572f671437152"}, "kind": "caption", "value": "Soft, faded frontal woman with shoulder-length brown hair on pale off-white background", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:05:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "5b36b3d40a979d8c"}
+{"target": {"record": "502caea572f671437152"}, "kind": "change", "value": "The pose becomes frontal with a different, softer woman, and the image is slightly hazy and washed out, with a pale warm-grey background.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:05:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a7535806cfb95899"}
+{"target": {"record": "502caea572f671437152"}, "kind": "keywords", "value": ["faded photo", "frontal portrait", "soft haze", "pale tones", "new identity"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:05:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "7da8a47f5cf1de90"}
+{"target": {"record": "502caea572f671437152"}, "kind": "effect_tags", "value": ["subject_kept", "washed_out", "blur"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:05:48Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ea99b20b272edfa2"}
diff --git a/records/sd1/sweep/502caea572f671437152/measurements.json b/records/sd1/sweep/502caea572f671437152/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..a0357f8c30f484782d14e06879aac552f4f078ab
--- /dev/null
+++ b/records/sd1/sweep/502caea572f671437152/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "502caea572f671437152",
+ "values": {
+  "cosine_distance": {
+   "value": 0.352366,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.451755,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.362256,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.150428,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.583,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00461,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 43.993,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0209,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0739,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/502caea572f671437152/record.json b/records/sd1/sweep/502caea572f671437152/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5692f0266815aa79a6e9fecd139d17845c85dcb7
--- /dev/null
+++ b/records/sd1/sweep/502caea572f671437152/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "502caea572f671437152",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:32Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.1.norm",
+   "op": "scale",
+   "args": {
+    "scale_factor": 2
+   },
+   "group": "in.hi",
+   "kind": "norm",
+   "module_type": "GroupNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "strong_grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/aad3c010ef58539bf2f680475fd05cf71fb7ff2f3771e493d457b1f10623d442.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "aad3c010ef58539bf2f680475fd05cf71fb7ff2f3771e493d457b1f10623d442",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "42c7eccc92662d396976"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/502caea572f671437152/workflow.json b/records/sd1/sweep/502caea572f671437152/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..17082351646c1b8b01f66f38ab0d1c34b2ce87f6
--- /dev/null
+++ b/records/sd1/sweep/502caea572f671437152/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 502caea572f671437152)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.1.norm\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50383a7bf3fb902769c0/measurements.json b/records/sd1/sweep/50383a7bf3fb902769c0/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..000182bf3da57940d8a0b0e7f3cea2370114af8a
--- /dev/null
+++ b/records/sd1/sweep/50383a7bf3fb902769c0/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "50383a7bf3fb902769c0",
+ "values": {
+  "cosine_distance": {
+   "value": 0.553054,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.688608,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.803132,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.427276,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 59.181,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0029,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 57.879,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8486,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2949,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50383a7bf3fb902769c0/record.json b/records/sd1/sweep/50383a7bf3fb902769c0/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..72cecffa55b113bb878e3c643dbcb9cfcd29a2c2
--- /dev/null
+++ b/records/sd1/sweep/50383a7bf3fb902769c0/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "50383a7bf3fb902769c0",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:15Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.0.skip_connection",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.mid",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/78b461f8f4971fb952ce1e1320b27a19331944ade2cfc74fcb1a6f931bc59650.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "78b461f8f4971fb952ce1e1320b27a19331944ade2cfc74fcb1a6f931bc59650",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "e6fa0b3c136e966edb9a"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50383a7bf3fb902769c0/workflow.json b/records/sd1/sweep/50383a7bf3fb902769c0/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..63ff32f03e2623ff4aa5e08ef1642cd37f65afda
--- /dev/null
+++ b/records/sd1/sweep/50383a7bf3fb902769c0/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50383a7bf3fb902769c0)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.0.skip_connection\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/503d987ecc45fe221fae/measurements.json b/records/sd1/sweep/503d987ecc45fe221fae/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cecd352e003a6d79ad278651762dd476282dd6e4
--- /dev/null
+++ b/records/sd1/sweep/503d987ecc45fe221fae/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "503d987ecc45fe221fae",
+ "values": {
+  "std": {
+   "value": 55.42,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01347,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 30.948,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.310429,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.475184,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.530461,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.108289,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9729,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1234,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/503d987ecc45fe221fae/record.json b/records/sd1/sweep/503d987ecc45fe221fae/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..acff88e62de18630fc6ae4703fb1bc61c0b262c4
--- /dev/null
+++ b/records/sd1/sweep/503d987ecc45fe221fae/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "503d987ecc45fe221fae",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:02:29Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 2.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "boost",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "13216623a742ccb32c5e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "multiply",
+   2.0,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e3c3cea080cfb5441c03"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/503d987ecc45fe221fae/workflow.json b/records/sd1/sweep/503d987ecc45fe221fae/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d9c9ec193013a2af42b9bef459f3ff4dd608919
--- /dev/null
+++ b/records/sd1/sweep/503d987ecc45fe221fae/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 503d987ecc45fe221fae)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/503ea2fa1e2e4770cd9f/measurements.json b/records/sd1/sweep/503ea2fa1e2e4770cd9f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2fe99b6d7703dcf757700bb4847c260d466dca34
--- /dev/null
+++ b/records/sd1/sweep/503ea2fa1e2e4770cd9f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "503ea2fa1e2e4770cd9f",
+ "values": {
+  "cosine_distance": {
+   "value": 0.082265,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.168882,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.086011,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.025328,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 58.958,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00718,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 14.623,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0421,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0601,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/503ea2fa1e2e4770cd9f/record.json b/records/sd1/sweep/503ea2fa1e2e4770cd9f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..da3c663f1389c13b8507ac56c67a6b754dda1c6d
--- /dev/null
+++ b/records/sd1/sweep/503ea2fa1e2e4770cd9f/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "503ea2fa1e2e4770cd9f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:18Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.0.out_layers.2",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/c8fec05b21ba40ef9de9c87ff0db251e59be750de87acb204d00f5ffbae818b0.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "c8fec05b21ba40ef9de9c87ff0db251e59be750de87acb204d00f5ffbae818b0",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "42f82da185f0a5a1a874"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/503ea2fa1e2e4770cd9f/workflow.json b/records/sd1/sweep/503ea2fa1e2e4770cd9f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a644f2d097eb4526a58025fb60e95531600fc98a
--- /dev/null
+++ b/records/sd1/sweep/503ea2fa1e2e4770cd9f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 503ea2fa1e2e4770cd9f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.0.out_layers.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/504d8f67d1184194384a/measurements.json b/records/sd1/sweep/504d8f67d1184194384a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..557f3cd8850f5fbb3fabc01caf93add4f17388d2
--- /dev/null
+++ b/records/sd1/sweep/504d8f67d1184194384a/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "504d8f67d1184194384a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.165241,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.22992,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.116062,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.035691,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 56.374,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00735,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 23.311,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9665,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1487,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/504d8f67d1184194384a/record.json b/records/sd1/sweep/504d8f67d1184194384a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9093522906c45bbdbbe15adf2e41ef28aca16b3f
--- /dev/null
+++ b/records/sd1/sweep/504d8f67d1184194384a/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "504d8f67d1184194384a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:40:26Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.1.transformer_blocks.0.attn1.to_q",
+   "op": "multiply",
+   "args": {
+    "scalar": 1.5
+   },
+   "group": "in.hi",
+   "kind": "self_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/723ee8cc7780bc78a1564877f4ab92d62514a634fa85cfa368ffcd94cbf12fa5.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "723ee8cc7780bc78a1564877f4ab92d62514a634fa85cfa368ffcd94cbf12fa5",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "3245f9472054cc7316de"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/504d8f67d1184194384a/workflow.json b/records/sd1/sweep/504d8f67d1184194384a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f32379833dc51ae5d42ae8138ca7e75197ab1116
--- /dev/null
+++ b/records/sd1/sweep/504d8f67d1184194384a/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 504d8f67d1184194384a)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.1.transformer_blocks.0.attn1.to_q\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505154907f65dea28c07/measurements.json b/records/sd1/sweep/505154907f65dea28c07/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..b680a97ee158ba51b494b13b745e1e065681a968
--- /dev/null
+++ b/records/sd1/sweep/505154907f65dea28c07/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "505154907f65dea28c07",
+ "values": {
+  "std": {
+   "value": 37.032,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00931,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 23.464,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.280832,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.312346,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.072804,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.068082,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0755,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1612,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505154907f65dea28c07/record.json b/records/sd1/sweep/505154907f65dea28c07/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..eed66723bc37c9131b9adaead60c938cb4cd00ad
--- /dev/null
+++ b/records/sd1/sweep/505154907f65dea28c07/record.json
@@ -0,0 +1,77 @@
+{
+ "id": "505154907f65dea28c07",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:20:38Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": -0.8
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "mid",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "2976534844fe80be26b4",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "add_scalar",
+   -0.8,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505154907f65dea28c07/workflow.json b/records/sd1/sweep/505154907f65dea28c07/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..10da4cec08674b994a4bbc2b788b29c1d5d4f9fb
--- /dev/null
+++ b/records/sd1/sweep/505154907f65dea28c07/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 505154907f65dea28c07)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": -0.8\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50532699d0fcfe95e60a/measurements.json b/records/sd1/sweep/50532699d0fcfe95e60a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c9aae47c37b955e3f8a9fe409924c6b3ff155757
--- /dev/null
+++ b/records/sd1/sweep/50532699d0fcfe95e60a/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "50532699d0fcfe95e60a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.190689,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.199392,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.183615,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.102529,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 58.665,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0033,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 23.405,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0275,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3155,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50532699d0fcfe95e60a/record.json b/records/sd1/sweep/50532699d0fcfe95e60a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..62f74af7ca3357f8f08832b3ec24492611796e8f
--- /dev/null
+++ b/records/sd1/sweep/50532699d0fcfe95e60a/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "50532699d0fcfe95e60a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:49Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.2.0.in_layers.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/ce1a2fdb0b37f2240b9860852944ad87eda849db288306c456633ed784711fdc.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "ce1a2fdb0b37f2240b9860852944ad87eda849db288306c456633ed784711fdc",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "017f20c663ec047a5df5"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50532699d0fcfe95e60a/workflow.json b/records/sd1/sweep/50532699d0fcfe95e60a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..22b2ff59ddae41eea87ff617cba572d0b5b6508d
--- /dev/null
+++ b/records/sd1/sweep/50532699d0fcfe95e60a/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50532699d0fcfe95e60a)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.2.0.in_layers.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505ae7d060ec0c6e9c4d/measurements.json b/records/sd1/sweep/505ae7d060ec0c6e9c4d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..be89fd80ccdddf35754e1f6788f6b2dff1f9c615
--- /dev/null
+++ b/records/sd1/sweep/505ae7d060ec0c6e9c4d/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "505ae7d060ec0c6e9c4d",
+ "values": {
+  "std": {
+   "value": 30.085,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02974,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.756,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.00012,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.000688,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.000861,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.000437,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0045,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2096,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505ae7d060ec0c6e9c4d/record.json b/records/sd1/sweep/505ae7d060ec0c6e9c4d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b804fdf7b4bf57ebe2bbe07c11439281f9a8eb98
--- /dev/null
+++ b/records/sd1/sweep/505ae7d060ec0c6e9c4d/record.json
@@ -0,0 +1,86 @@
+{
+ "id": "505ae7d060ec0c6e9c4d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:05:29Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.1",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.2,
+    "seed": 7
+   },
+   "group": "out.lo",
+   "kind": "attn_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "medium",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.lo",
+   "add_noise",
+   0.2,
+   "late"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.2,
+    "applied": 0.2,
+    "reason": "outside safe range [0.3, 3.0] for out.lo (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "37b11192a3a64a62c056"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505ae7d060ec0c6e9c4d/workflow.json b/records/sd1/sweep/505ae7d060ec0c6e9c4d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..50251dc3e0f939ad5a3b16432e0e88c0effbeb12
--- /dev/null
+++ b/records/sd1/sweep/505ae7d060ec0c6e9c4d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 505ae7d060ec0c6e9c4d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.1\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.2,\n    \"seed\": 7\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505c97535b86671e7479/measurements.json b/records/sd1/sweep/505c97535b86671e7479/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..1f3726c712df38459d9bffc8ecdc570edaedd200
--- /dev/null
+++ b/records/sd1/sweep/505c97535b86671e7479/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "505c97535b86671e7479",
+ "values": {
+  "cosine_distance": {
+   "value": 0.306989,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.303411,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.166982,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.104133,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 56.52,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00734,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 30.632,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9631,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4858,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505c97535b86671e7479/record.json b/records/sd1/sweep/505c97535b86671e7479/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..fb5d070ba322e30629a68d664be38b3f6a5b6482
--- /dev/null
+++ b/records/sd1/sweep/505c97535b86671e7479/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "505c97535b86671e7479",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:03Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.2.0.out_layers.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.5
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/6978b71e14b547b407e1ba685460cabbaca3c17a281ff547c215a52b5a186760.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "6978b71e14b547b407e1ba685460cabbaca3c17a281ff547c215a52b5a186760",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "aeb54686a8dd0eb8d4a8"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/505c97535b86671e7479/workflow.json b/records/sd1/sweep/505c97535b86671e7479/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f99546bc0558dbb2669d87507970089369f23c2e
--- /dev/null
+++ b/records/sd1/sweep/505c97535b86671e7479/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 505c97535b86671e7479)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.2.0.out_layers.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50682029ce6215e6189f/measurements.json b/records/sd1/sweep/50682029ce6215e6189f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2fdd04c32d80fe59044617d330930e5cbbad4142
--- /dev/null
+++ b/records/sd1/sweep/50682029ce6215e6189f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50682029ce6215e6189f",
+ "values": {
+  "std": {
+   "value": 44.168,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0152,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 7.08,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.017575,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.055473,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.015673,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.007344,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0063,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0317,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50682029ce6215e6189f/record.json b/records/sd1/sweep/50682029ce6215e6189f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..65b8dc9596397d85389efd95d5cbac3bfbbe9082
--- /dev/null
+++ b/records/sd1/sweep/50682029ce6215e6189f/record.json
@@ -0,0 +1,79 @@
+{
+ "id": "50682029ce6215e6189f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:00:16Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.8,
+    "seed": 7
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "heavy",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "add_noise",
+   0.8,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "07f3412072aa4c60b635"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50682029ce6215e6189f/workflow.json b/records/sd1/sweep/50682029ce6215e6189f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7acd4444a6ead2cc2c65b3a37563f66088fb0993
--- /dev/null
+++ b/records/sd1/sweep/50682029ce6215e6189f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50682029ce6215e6189f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.8,\n    \"seed\": 7\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506838c2e54e982984e1/interpretations.jsonl b/records/sd1/sweep/506838c2e54e982984e1/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..0b611d6fa84827dff66cc9907851a616c2e11fa5
--- /dev/null
+++ b/records/sd1/sweep/506838c2e54e982984e1/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "506838c2e54e982984e1"}, "kind": "caption", "value": "Painted woman with wavy dark hair and a black turtleneck on a grey textured panel", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:23:30Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "6c30d06234b4fdd9"}
+{"target": {"record": "506838c2e54e982984e1"}, "kind": "change", "value": "Almost no visible change; the face is a little more mature and the border slightly narrower.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:23:30Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "8e61616bef80be0f"}
+{"target": {"record": "506838c2e54e982984e1"}, "kind": "keywords", "value": ["portrait", "minor variation", "turtleneck", "painted portrait"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:23:30Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "2956db3f6e5ec74a"}
+{"target": {"record": "506838c2e54e982984e1"}, "kind": "effect_tags", "value": ["subtle", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:23:30Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "da5bd561b152c31a"}
diff --git a/records/sd1/sweep/506838c2e54e982984e1/measurements.json b/records/sd1/sweep/506838c2e54e982984e1/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..438e22ff115a1207109a8f998717909bd1087c7c
--- /dev/null
+++ b/records/sd1/sweep/506838c2e54e982984e1/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "506838c2e54e982984e1",
+ "values": {
+  "cosine_distance": {
+   "value": 0.028909,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.047197,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.010329,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.006184,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.496,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00599,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 7.039,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0323,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1007,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506838c2e54e982984e1/record.json b/records/sd1/sweep/506838c2e54e982984e1/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..92871a52935e266801ad99de99610206b6029d11
--- /dev/null
+++ b/records/sd1/sweep/506838c2e54e982984e1/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "506838c2e54e982984e1",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:43:28Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "middle_block.2.in_layers.2",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/0b966d58a7fe0586ec992f86984632a13e7211d7f2864c65e52b28b32305ca94.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "0b966d58a7fe0586ec992f86984632a13e7211d7f2864c65e52b28b32305ca94",
+  "type_path": "UNetModel.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "4f2926f65ca0940b249a"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506838c2e54e982984e1/workflow.json b/records/sd1/sweep/506838c2e54e982984e1/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..2ed949efb5958bc915d402073112660293e0a3a1
--- /dev/null
+++ b/records/sd1/sweep/506838c2e54e982984e1/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 506838c2e54e982984e1)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.2.in_layers.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506d16eec41cffebdda9/interpretations.jsonl b/records/sd1/sweep/506d16eec41cffebdda9/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..b1c3222d0b79664c28c7820dc7ea64436057d4dd
--- /dev/null
+++ b/records/sd1/sweep/506d16eec41cffebdda9/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "506d16eec41cffebdda9"}, "kind": "caption", "value": "Small inverted figure with long hair beneath a symmetrical teal and copper geometric structure", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:08:15Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "eba5bb569c32e1c2"}
+{"target": {"record": "506d16eec41cffebdda9"}, "kind": "change", "value": "The portrait breaks into a symmetrical, kaleidoscopic architectural pattern of teal and copper panels, with a tiny, upside-down distorted figure at the bottom.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:08:15Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ea8cf17e4592f5fe"}
+{"target": {"record": "506d16eec41cffebdda9"}, "kind": "keywords", "value": ["kaleidoscope", "geometric panels", "inverted figure", "symmetry", "teal copper"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:08:15Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "191abea6455c02cd"}
+{"target": {"record": "506d16eec41cffebdda9"}, "kind": "effect_tags", "value": ["abstract", "geometric", "symmetry", "surreal", "subject_lost", "new_composition", "distorted"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:08:15Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "854cb3d789957e40"}
diff --git a/records/sd1/sweep/506d16eec41cffebdda9/measurements.json b/records/sd1/sweep/506d16eec41cffebdda9/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..25d4f16a8bed8bc2a9eec05c7462a176e0b18ace
--- /dev/null
+++ b/records/sd1/sweep/506d16eec41cffebdda9/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "506d16eec41cffebdda9",
+ "values": {
+  "cosine_distance": {
+   "value": 0.591956,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.638307,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.825754,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.443654,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 64.8,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00963,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 56.684,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.7359,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.6768,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506d16eec41cffebdda9/record.json b/records/sd1/sweep/506d16eec41cffebdda9/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..8093ae111b93f15b8075ad73b3f48dcdbbb91fdc
--- /dev/null
+++ b/records/sd1/sweep/506d16eec41cffebdda9/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "506d16eec41cffebdda9",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:48:54Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.1.transformer_blocks.0.ff.net.1",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "in.lo",
+   "kind": "ff",
+   "module_type": "Dropout",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/5811e048977aeb41697670c4f34dd79c51b8dfa533bd28b81b38998c796240c3.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "5811e048977aeb41697670c4f34dd79c51b8dfa533bd28b81b38998c796240c3",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.FeedForward.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "a8bd9edd77c14415ad44"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506d16eec41cffebdda9/workflow.json b/records/sd1/sweep/506d16eec41cffebdda9/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f6bc8f032c84dccd16c60e33b41aa101d8cbcd92
--- /dev/null
+++ b/records/sd1/sweep/506d16eec41cffebdda9/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 506d16eec41cffebdda9)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.1.transformer_blocks.0.ff.net.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506fd1c8331e0598b997/measurements.json b/records/sd1/sweep/506fd1c8331e0598b997/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..8881dc102567ec80f0a7f128ba2bd595e227d23f
--- /dev/null
+++ b/records/sd1/sweep/506fd1c8331e0598b997/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "506fd1c8331e0598b997",
+ "values": {
+  "std": {
+   "value": 72.729,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00568,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 61.465,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.618458,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.638566,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.960928,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.456103,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.6432,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7057,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506fd1c8331e0598b997/record.json b/records/sd1/sweep/506fd1c8331e0598b997/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..e97eb8103b29153f477437a406a9c5374636223b
--- /dev/null
+++ b/records/sd1/sweep/506fd1c8331e0598b997/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "506fd1c8331e0598b997",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:30:57Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30.0
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "slight",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "2976534844fe80be26b4",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "rotate",
+   30.0,
+   "middle"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 30.0,
+    "applied": 30.0,
+    "reason": "outside safe range [0.0, 5.0] for out.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "842483f4db13c2fd9e34"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/506fd1c8331e0598b997/workflow.json b/records/sd1/sweep/506fd1c8331e0598b997/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..5b739757c9a4c760e093ff14ee5670ddf989506c
--- /dev/null
+++ b/records/sd1/sweep/506fd1c8331e0598b997/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 506fd1c8331e0598b997)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50741e106e57b80d32a5/interpretations.jsonl b/records/sd1/sweep/50741e106e57b80d32a5/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..7b76b75f23426a4ee5f30df33f430365b4551d2e
--- /dev/null
+++ b/records/sd1/sweep/50741e106e57b80d32a5/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "50741e106e57b80d32a5"}, "kind": "caption", "value": "Faded, smeared lighthouse on rocky coast at sunset with ghostly, washed-out tower.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:45:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "b427627ccc4228c9"}
+{"target": {"record": "50741e106e57b80d32a5"}, "kind": "change", "value": "Same composition but the lighthouse and coast blur into a hazy, washed-out, painterly smear with muted colours.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:45:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "7495866d67821c50"}
+{"target": {"record": "50741e106e57b80d32a5"}, "kind": "keywords", "value": ["faded lighthouse", "smeared painting", "hazy coast", "washed out sunset"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:45:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "7f9d790bff2926ac"}
+{"target": {"record": "50741e106e57b80d32a5"}, "kind": "effect_tags", "value": ["washed_out", "blur", "painterly", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:45:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "bcace06b3f7b564d"}
diff --git a/records/sd1/sweep/50741e106e57b80d32a5/measurements.json b/records/sd1/sweep/50741e106e57b80d32a5/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..ce60a8d52b0912bdacd49de57d1202f94c31abd4
--- /dev/null
+++ b/records/sd1/sweep/50741e106e57b80d32a5/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50741e106e57b80d32a5",
+ "values": {
+  "std": {
+   "value": 43.757,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01542,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 13.126,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.06653,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.229008,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.35825,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.138166,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9403,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1642,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50741e106e57b80d32a5/record.json b/records/sd1/sweep/50741e106e57b80d32a5/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..4840559010efc25151b8e4da78e609145ab34341
--- /dev/null
+++ b/records/sd1/sweep/50741e106e57b80d32a5/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "50741e106e57b80d32a5",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:50:52Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "zero",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "multiply",
+   0.0,
+   "late"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for in.lo (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "164e3d794948259aec0d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50741e106e57b80d32a5/workflow.json b/records/sd1/sweep/50741e106e57b80d32a5/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..abada07d0dec8743bdec9e09d767da4c6a7871fb
--- /dev/null
+++ b/records/sd1/sweep/50741e106e57b80d32a5/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50741e106e57b80d32a5)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5081b86bcd696a568d76/interpretations.jsonl b/records/sd1/sweep/5081b86bcd696a568d76/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..47041d31aefd3e0ddc4479fb80efdb38529fd47d
--- /dev/null
+++ b/records/sd1/sweep/5081b86bcd696a568d76/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "5081b86bcd696a568d76"}, "kind": "caption", "value": "Extreme close-up of a blue eye, fringe and rosy cheek of a young woman", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:19:23Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "8b5e84a28a5e5e98"}
+{"target": {"record": "5081b86bcd696a568d76"}, "kind": "change", "value": "The camera moves into an extreme close-up of one eye and cheek, with warm skin tones, a crisp blue iris and a cropped fringe.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:19:23Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "e6a408343a137fe7"}
+{"target": {"record": "5081b86bcd696a568d76"}, "kind": "keywords", "value": ["extreme close-up", "blue eye", "macro portrait", "fringe", "rosy cheek"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:19:23Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "696219aefe207f1e"}
+{"target": {"record": "5081b86bcd696a568d76"}, "kind": "effect_tags", "value": ["zoom", "fine_detail", "new_composition", "colour_shift"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:19:23Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "aa3416b3d14c1dc2"}
diff --git a/records/sd1/sweep/5081b86bcd696a568d76/measurements.json b/records/sd1/sweep/5081b86bcd696a568d76/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c712a0db211894fbed30e3bf611a9582402f4202
--- /dev/null
+++ b/records/sd1/sweep/5081b86bcd696a568d76/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5081b86bcd696a568d76",
+ "values": {
+  "cosine_distance": {
+   "value": 0.522167,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.65457,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.453286,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.271286,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 62.609,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00637,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 53.355,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9177,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.727,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5081b86bcd696a568d76/record.json b/records/sd1/sweep/5081b86bcd696a568d76/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..e020f4f26e9347090b25698ea5ef6ff2016a89bf
--- /dev/null
+++ b/records/sd1/sweep/5081b86bcd696a568d76/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "5081b86bcd696a568d76",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:00Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.0.0.out_layers.2",
+   "op": "scale",
+   "args": {
+    "scale_factor": 2
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "strong_grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/f07fe30d036e738f84f3ce843ad44330da8bfdf511c8ab0992fa1b4cb769133c.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "f07fe30d036e738f84f3ce843ad44330da8bfdf511c8ab0992fa1b4cb769133c",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "76b492bfdfcc774c1ebc"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5081b86bcd696a568d76/workflow.json b/records/sd1/sweep/5081b86bcd696a568d76/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..20bbe01533111db6894e584d2c83d589f56ff380
--- /dev/null
+++ b/records/sd1/sweep/5081b86bcd696a568d76/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5081b86bcd696a568d76)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.0.0.out_layers.2\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/508c7070a51843c33329/measurements.json b/records/sd1/sweep/508c7070a51843c33329/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..8fdc1eeb2ea8224368938387c1375f83470a991e
--- /dev/null
+++ b/records/sd1/sweep/508c7070a51843c33329/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "508c7070a51843c33329",
+ "values": {
+  "std": {
+   "value": 53.957,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01385,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 29.362,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.278337,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.372433,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.1882,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.05675,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9103,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.545,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/508c7070a51843c33329/record.json b/records/sd1/sweep/508c7070a51843c33329/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..8a1407dd112935adbd1284927c83469ed55e9e4f
--- /dev/null
+++ b/records/sd1/sweep/508c7070a51843c33329/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "508c7070a51843c33329",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:07:02Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 2.0
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "boost",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "13216623a742ccb32c5e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "multiply",
+   2.0,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6beaa2fcf4ed2f591274"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/508c7070a51843c33329/workflow.json b/records/sd1/sweep/508c7070a51843c33329/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ec28dfa28d8dbcf5e75c3b5dbf02d080c6982734
--- /dev/null
+++ b/records/sd1/sweep/508c7070a51843c33329/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 508c7070a51843c33329)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5095d9f6b878f4c52aeb/measurements.json b/records/sd1/sweep/5095d9f6b878f4c52aeb/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..470aae65732ae71f93a7cda115463ce828655366
--- /dev/null
+++ b/records/sd1/sweep/5095d9f6b878f4c52aeb/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5095d9f6b878f4c52aeb",
+ "values": {
+  "cosine_distance": {
+   "value": 0.109118,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.147799,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.128551,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.040429,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 59.072,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00529,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 15.289,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0455,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.042,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5095d9f6b878f4c52aeb/record.json b/records/sd1/sweep/5095d9f6b878f4c52aeb/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..725e9ddae4670457a103caf982c48cbed01f167f
--- /dev/null
+++ b/records/sd1/sweep/5095d9f6b878f4c52aeb/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5095d9f6b878f4c52aeb",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:44:28Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.1.transformer_blocks.0.attn2.to_out.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.lo",
+   "kind": "cross_attn",
+   "module_type": "Dropout",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/cd76fe97a4e33214b72bc6476858b94d8ba5d1155b481571976af469325c7fba.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "cd76fe97a4e33214b72bc6476858b94d8ba5d1155b481571976af469325c7fba",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "53714f15dd7c79214dc5"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5095d9f6b878f4c52aeb/workflow.json b/records/sd1/sweep/5095d9f6b878f4c52aeb/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e81467389945a2993b47eef5ddeff3818ef5d30
--- /dev/null
+++ b/records/sd1/sweep/5095d9f6b878f4c52aeb/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5095d9f6b878f4c52aeb)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.1.transformer_blocks.0.attn2.to_out.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5096be73f3eb141aba7e/measurements.json b/records/sd1/sweep/5096be73f3eb141aba7e/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2dbb556df9372f6b664399f346406d709f39f30d
--- /dev/null
+++ b/records/sd1/sweep/5096be73f3eb141aba7e/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5096be73f3eb141aba7e",
+ "values": {
+  "cosine_distance": {
+   "value": 0.210229,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.302712,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.212483,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.093008,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 56.824,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00438,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 28.75,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.957,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0255,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5096be73f3eb141aba7e/record.json b/records/sd1/sweep/5096be73f3eb141aba7e/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..983f969df16fe9ae13b2e7d54e1c4003c12d2dc3
--- /dev/null
+++ b/records/sd1/sweep/5096be73f3eb141aba7e/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5096be73f3eb141aba7e",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:06Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.4.0.out_layers.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "in.mid",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/10d8680921bfb12adf3e3cd586840131c54270424cdf0e57cfc1f8a53252c725.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "10d8680921bfb12adf3e3cd586840131c54270424cdf0e57cfc1f8a53252c725",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "c5ddc28207389c264a1f"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5096be73f3eb141aba7e/workflow.json b/records/sd1/sweep/5096be73f3eb141aba7e/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..be97770cae2d3f9782a53a508ebec6f3fc9730e2
--- /dev/null
+++ b/records/sd1/sweep/5096be73f3eb141aba7e/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5096be73f3eb141aba7e)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.4.0.out_layers.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/509a9122e5749692552a/interpretations.jsonl b/records/sd1/sweep/509a9122e5749692552a/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..0af5ee5508455d33cb158fdf7c99a2397f09f3ec
--- /dev/null
+++ b/records/sd1/sweep/509a9122e5749692552a/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "509a9122e5749692552a"}, "kind": "caption", "value": "Abstract camouflage-like texture of beige and navy brushstrokes with faint face ghost", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:34:37Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "569e95288bdecca8"}
+{"target": {"record": "509a9122e5749692552a"}, "kind": "change", "value": "The portrait collapses into a dense mottled abstract texture of navy, beige and grey strokes with only a faint oval hint of the face.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:34:37Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "eb6bbd65e46e2dc9"}
+{"target": {"record": "509a9122e5749692552a"}, "kind": "keywords", "value": ["abstract texture", "camouflage pattern", "dissolved portrait", "navy and beige", "brushstroke noise"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:34:37Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "15288cb0fe8d4dd0"}
+{"target": {"record": "509a9122e5749692552a"}, "kind": "effect_tags", "value": ["abstract", "subject_lost", "texture", "pattern", "fragmented", "degenerate"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:34:37Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "4ecc0d42b1ebbc5d"}
diff --git a/records/sd1/sweep/509a9122e5749692552a/measurements.json b/records/sd1/sweep/509a9122e5749692552a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..61aa544d014dec395e2d2eb858c4a1b04b80a79c
--- /dev/null
+++ b/records/sd1/sweep/509a9122e5749692552a/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "509a9122e5749692552a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.443045,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.611626,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.805196,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.397212,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 45.148,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01353,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 48.409,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8394,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1168,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/509a9122e5749692552a/record.json b/records/sd1/sweep/509a9122e5749692552a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b741e846ae864d1798b7fc541db9d452b403f7c
--- /dev/null
+++ b/records/sd1/sweep/509a9122e5749692552a/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "509a9122e5749692552a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:19Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.0.skip_connection",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "out.mid",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/580ef29e664ed4a74abbdc06af27db16988eabf7d0c703a80c6658c1531aa3e8.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "580ef29e664ed4a74abbdc06af27db16988eabf7d0c703a80c6658c1531aa3e8",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "837d7e2b22283f9f311e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/509a9122e5749692552a/workflow.json b/records/sd1/sweep/509a9122e5749692552a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..cc249fa428db9d5c3313f65852153fc18ff6272a
--- /dev/null
+++ b/records/sd1/sweep/509a9122e5749692552a/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 509a9122e5749692552a)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.0.skip_connection\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/509c6b6e8779c39e49ff/measurements.json b/records/sd1/sweep/509c6b6e8779c39e49ff/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2ceea112c5fa5074ed11a4ef76ce02b2091947bd
--- /dev/null
+++ b/records/sd1/sweep/509c6b6e8779c39e49ff/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "509c6b6e8779c39e49ff",
+ "values": {
+  "cosine_distance": {
+   "value": 0.158017,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.170807,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.188125,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.131924,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.889,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00561,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 17.678,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9726,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4369,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/509c6b6e8779c39e49ff/record.json b/records/sd1/sweep/509c6b6e8779c39e49ff/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..28e067ace7df661abb31a87d0ca7bc368bac4f9e
--- /dev/null
+++ b/records/sd1/sweep/509c6b6e8779c39e49ff/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "509c6b6e8779c39e49ff",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:42Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "middle_block.2.out_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "mid",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/fb47f4ab140aaac1852eafc810ff7fbf18461e601c6dafb2729dc001fb1f4aeb.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "fb47f4ab140aaac1852eafc810ff7fbf18461e601c6dafb2729dc001fb1f4aeb",
+  "type_path": "UNetModel.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "1b3f5d76f0b0d8ffe3f6"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/509c6b6e8779c39e49ff/workflow.json b/records/sd1/sweep/509c6b6e8779c39e49ff/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..4c340e105ea59125eed6d3092399fce1651d231b
--- /dev/null
+++ b/records/sd1/sweep/509c6b6e8779c39e49ff/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 509c6b6e8779c39e49ff)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.2.out_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50ab9281800acc460ee9/interpretations.jsonl b/records/sd1/sweep/50ab9281800acc460ee9/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..ea9abf934927406b4d29bc8ed9c1a867cecc6bc0
--- /dev/null
+++ b/records/sd1/sweep/50ab9281800acc460ee9/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "50ab9281800acc460ee9"}, "kind": "caption", "value": "Daytime lighthouse with red-roofed keeper's house on rocky islet under blue sky and cumulus clouds", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:47:47Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "2bd6a493cd0a35d8"}
+{"target": {"record": "50ab9281800acc460ee9"}, "kind": "change", "value": "The sunset scene becomes daytime: warm orange sky turns to blue with white cumulus clouds, a keeper's house is added beside the lighthouse, and the island composition is reworked.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:47:47Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "2a2e5ea34c66b837"}
+{"target": {"record": "50ab9281800acc460ee9"}, "kind": "keywords", "value": ["sunset to daytime", "blue sky", "lighthouse cottage", "cumulus clouds", "cooler palette"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:47:47Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "f1c51e5e50d97b95"}
+{"target": {"record": "50ab9281800acc460ee9"}, "kind": "effect_tags", "value": ["colour_shift", "bright", "new_composition", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:47:47Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "41501b6108cac829"}
diff --git a/records/sd1/sweep/50ab9281800acc460ee9/measurements.json b/records/sd1/sweep/50ab9281800acc460ee9/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0638b73aee5a976f384a4060664da04836282174
--- /dev/null
+++ b/records/sd1/sweep/50ab9281800acc460ee9/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50ab9281800acc460ee9",
+ "values": {
+  "std": {
+   "value": 46.265,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01854,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 41.566,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.485073,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.545121,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.135842,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.059217,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.023,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0077,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50ab9281800acc460ee9/record.json b/records/sd1/sweep/50ab9281800acc460ee9/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a381685ac1368c0be437650978d656da3e00edcd
--- /dev/null
+++ b/records/sd1/sweep/50ab9281800acc460ee9/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "50ab9281800acc460ee9",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:02:35Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "zero",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "multiply",
+   0.0,
+   "early"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "b74b194fffd298180125"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50ab9281800acc460ee9/workflow.json b/records/sd1/sweep/50ab9281800acc460ee9/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..1ad39a92b86da88a7504d002cf40612a79df4186
--- /dev/null
+++ b/records/sd1/sweep/50ab9281800acc460ee9/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50ab9281800acc460ee9)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b06ca31eb4d7dc0566/measurements.json b/records/sd1/sweep/50b06ca31eb4d7dc0566/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..074baf3f093194fd2eabf86d301a5ef437d93cce
--- /dev/null
+++ b/records/sd1/sweep/50b06ca31eb4d7dc0566/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "50b06ca31eb4d7dc0566",
+ "values": {
+  "cosine_distance": {
+   "value": 0.046114,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.057891,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.026798,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.010455,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 54.421,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00492,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 8.676,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9831,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1678,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b06ca31eb4d7dc0566/record.json b/records/sd1/sweep/50b06ca31eb4d7dc0566/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..939d0ad15d974dfb2152cb326b6687449b931b00
--- /dev/null
+++ b/records/sd1/sweep/50b06ca31eb4d7dc0566/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "50b06ca31eb4d7dc0566",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:39:56Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.1.transformer_blocks.0.attn2.to_v",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.hi",
+   "kind": "cross_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/b37ce0cdacc129a87cb82ec9c6e06119dbae36d9aa4d56a83d7d8eff67b4c64a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "b37ce0cdacc129a87cb82ec9c6e06119dbae36d9aa4d56a83d7d8eff67b4c64a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a36ef8ddd30703539718"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b06ca31eb4d7dc0566/workflow.json b/records/sd1/sweep/50b06ca31eb4d7dc0566/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ece38eb67ae8bc5704196604d154f6dd16a1e94f
--- /dev/null
+++ b/records/sd1/sweep/50b06ca31eb4d7dc0566/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50b06ca31eb4d7dc0566)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.1.transformer_blocks.0.attn2.to_v\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b1511bc1e4725638ec/measurements.json b/records/sd1/sweep/50b1511bc1e4725638ec/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..fa8aed59fb11e63ace8541c4ec4bd6e4c8eb8e92
--- /dev/null
+++ b/records/sd1/sweep/50b1511bc1e4725638ec/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "50b1511bc1e4725638ec",
+ "values": {
+  "cosine_distance": {
+   "value": 0.102367,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.124704,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.186839,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.030388,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 64.858,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00541,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 14.819,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0075,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1692,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b1511bc1e4725638ec/record.json b/records/sd1/sweep/50b1511bc1e4725638ec/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..eaae9acbc0396e2482c92297cb68f3af935c6222
--- /dev/null
+++ b/records/sd1/sweep/50b1511bc1e4725638ec/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "50b1511bc1e4725638ec",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:08Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.5.0.out_layers.3",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/ee2bde3e05470e2f4d82c2c6652e2ffee64a12ce5b09936218339ae451d3ce87.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "ee2bde3e05470e2f4d82c2c6652e2ffee64a12ce5b09936218339ae451d3ce87",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "0db4c4a71049124d9e47"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b1511bc1e4725638ec/workflow.json b/records/sd1/sweep/50b1511bc1e4725638ec/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..1693ecfd5c34f8cc653a22d0fbc03aa639acfd70
--- /dev/null
+++ b/records/sd1/sweep/50b1511bc1e4725638ec/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50b1511bc1e4725638ec)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.5.0.out_layers.3\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b3b929270998c73053/measurements.json b/records/sd1/sweep/50b3b929270998c73053/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..04d715771df7e9982dc0c70596eb50d6c580f260
--- /dev/null
+++ b/records/sd1/sweep/50b3b929270998c73053/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "50b3b929270998c73053",
+ "values": {
+  "cosine_distance": {
+   "value": 0.271883,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.343284,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.246381,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.126441,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 64.301,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00304,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 34.36,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9706,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2825,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b3b929270998c73053/record.json b/records/sd1/sweep/50b3b929270998c73053/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..c521afd8fa67fc4a222dc0548c6cb259d8df0e72
--- /dev/null
+++ b/records/sd1/sweep/50b3b929270998c73053/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "50b3b929270998c73053",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:39Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0.in_layers.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.5
+   },
+   "group": "in.mid",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/7b9685b598c58b436190dc71ac0395c3e109b289967a1863ae2144d22b5977b1.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "7b9685b598c58b436190dc71ac0395c3e109b289967a1863ae2144d22b5977b1",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "8700892f8364c81c2542"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50b3b929270998c73053/workflow.json b/records/sd1/sweep/50b3b929270998c73053/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..5eee1b22539027617e13a248a7489659b48ef2ca
--- /dev/null
+++ b/records/sd1/sweep/50b3b929270998c73053/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50b3b929270998c73053)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0.in_layers.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50daf1991a215b097a32/measurements.json b/records/sd1/sweep/50daf1991a215b097a32/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..811db5f3c836b05398dee58431180fa12e1ab477
--- /dev/null
+++ b/records/sd1/sweep/50daf1991a215b097a32/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "50daf1991a215b097a32",
+ "values": {
+  "cosine_distance": {
+   "value": 0.17304,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.27563,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.263537,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.165939,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 68.826,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00177,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 22.702,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9981,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1296,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50daf1991a215b097a32/record.json b/records/sd1/sweep/50daf1991a215b097a32/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1b4fade7d537bd43b65e56161e68b228f533ff8f
--- /dev/null
+++ b/records/sd1/sweep/50daf1991a215b097a32/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "50daf1991a215b097a32",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:26Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.11.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/4a22db4aa236540144b5b7d9c43fce2f1c5b20af3514c6ae23cb066ba37b891d.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "4a22db4aa236540144b5b7d9c43fce2f1c5b20af3514c6ae23cb066ba37b891d",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "5dac9cc35c2664b9dd70"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50daf1991a215b097a32/workflow.json b/records/sd1/sweep/50daf1991a215b097a32/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..66a28cf3510c514f63a6ad4d377440549268fbf9
--- /dev/null
+++ b/records/sd1/sweep/50daf1991a215b097a32/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50daf1991a215b097a32)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.11.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50e51ff2533aa7331209/measurements.json b/records/sd1/sweep/50e51ff2533aa7331209/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cc53100483e6e7cc43df7c7fecfb486954ed15e4
--- /dev/null
+++ b/records/sd1/sweep/50e51ff2533aa7331209/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50e51ff2533aa7331209",
+ "values": {
+  "cosine_distance": {
+   "value": 0.266953,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.392322,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.264922,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.089204,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 48.859,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0051,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 30.362,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9879,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0359,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50e51ff2533aa7331209/record.json b/records/sd1/sweep/50e51ff2533aa7331209/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..312bb13dc9d989e1ac49dbab2ac33e7ad1dbe6b3
--- /dev/null
+++ b/records/sd1/sweep/50e51ff2533aa7331209/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "50e51ff2533aa7331209",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:17Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.0.out_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/8969f6b13f1afc19b95588410afc1cff43f52f291708ee0a121389a8e9a6cb27.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "8969f6b13f1afc19b95588410afc1cff43f52f291708ee0a121389a8e9a6cb27",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6911753592f97002836d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50e51ff2533aa7331209/workflow.json b/records/sd1/sweep/50e51ff2533aa7331209/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d9e9aad22ee5a8712c453be76bc2215c83a9b4b
--- /dev/null
+++ b/records/sd1/sweep/50e51ff2533aa7331209/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50e51ff2533aa7331209)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.0.out_layers.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50e96eac059bdd1ffb94/measurements.json b/records/sd1/sweep/50e96eac059bdd1ffb94/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..4db3de4b85474dc573465d189a5c282ea15dc75d
--- /dev/null
+++ b/records/sd1/sweep/50e96eac059bdd1ffb94/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50e96eac059bdd1ffb94",
+ "values": {
+  "std": {
+   "value": 40.842,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0125,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 32.189,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.341515,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.601806,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.831548,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.305901,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9385,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.027,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50e96eac059bdd1ffb94/record.json b/records/sd1/sweep/50e96eac059bdd1ffb94/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d8f96d82e2ec3e775f3b74b70c85e000f2b4a7ea
--- /dev/null
+++ b/records/sd1/sweep/50e96eac059bdd1ffb94/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "50e96eac059bdd1ffb94",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:49:26Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "rotate",
+   90.0,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 90.0,
+    "applied": 90.0,
+    "reason": "outside safe range [5.0, 45.0] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "ae7a740a55903a80dbd2"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50e96eac059bdd1ffb94/workflow.json b/records/sd1/sweep/50e96eac059bdd1ffb94/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7a994e91eead7bfb9309ac0ed3d37d2576d2d106
--- /dev/null
+++ b/records/sd1/sweep/50e96eac059bdd1ffb94/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50e96eac059bdd1ffb94)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90.0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50f45e9536cba5937708/interpretations.jsonl b/records/sd1/sweep/50f45e9536cba5937708/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..45f9af62091cd259c8497503773915469e38961f
--- /dev/null
+++ b/records/sd1/sweep/50f45e9536cba5937708/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "50f45e9536cba5937708"}, "kind": "caption", "value": "Loose oil portrait of a woman with a dark choppy bob on pale grey canvas", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:14:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "6c268c2d7f1dfd54"}
+{"target": {"record": "50f45e9536cba5937708"}, "kind": "change", "value": "The smooth illustration becomes a rough oil painting: the hair turns into a darker, shorter, choppy bob, the brushstrokes show, and the dark background and paper border become pale streaky grey.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:14:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "079f7e6aa8538e93"}
+{"target": {"record": "50f45e9536cba5937708"}, "kind": "keywords", "value": ["oil painting", "choppy bob", "visible brushstrokes", "grey background", "portrait"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:14:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "af06a6febe038af4"}
+{"target": {"record": "50f45e9536cba5937708"}, "kind": "effect_tags", "value": ["painterly", "style_change", "texture", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:14:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "8414d9e142c6378b"}
diff --git a/records/sd1/sweep/50f45e9536cba5937708/measurements.json b/records/sd1/sweep/50f45e9536cba5937708/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0762762954338788ce854863646ecb542340563c
--- /dev/null
+++ b/records/sd1/sweep/50f45e9536cba5937708/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "50f45e9536cba5937708",
+ "values": {
+  "cosine_distance": {
+   "value": 0.282387,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.343713,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.260283,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.075216,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.258,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0061,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 35.988,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0162,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0362,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50f45e9536cba5937708/record.json b/records/sd1/sweep/50f45e9536cba5937708/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..95170d4e756657c368f42bf39958aa18d0f9996c
--- /dev/null
+++ b/records/sd1/sweep/50f45e9536cba5937708/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "50f45e9536cba5937708",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:00Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.1.proj_in",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "in.hi",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/519a2babebdff47abe11073c9d5a0a280336092f1349405eece499adc777c63a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "519a2babebdff47abe11073c9d5a0a280336092f1349405eece499adc777c63a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "72aa4e554a5da448401e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50f45e9536cba5937708/workflow.json b/records/sd1/sweep/50f45e9536cba5937708/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c61c175cd434e09d21acd607c61005f85a147cb1
--- /dev/null
+++ b/records/sd1/sweep/50f45e9536cba5937708/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50f45e9536cba5937708)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.1.proj_in\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/measurements.json b/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..77760391186880b7be039f2f32d6554e5b2c8ad3
--- /dev/null
+++ b/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "50fd72d0f9c9b9b8ce7d",
+ "values": {
+  "cosine_distance": {
+   "value": 0.541177,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.654697,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.839731,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.350682,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 64.223,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00285,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 51.797,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9881,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4688,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/record.json b/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..6845032871dfa1817191dc5ce5f2fb264a96e366
--- /dev/null
+++ b/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "50fd72d0f9c9b9b8ce7d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:47:58Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0.skip_connection",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "in.hi",
+   "kind": "skip",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/c93beb5534ef956ba5724ed0a7d35cbf131e29e13e44280180b44e3993bde095.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "c93beb5534ef956ba5724ed0a7d35cbf131e29e13e44280180b44e3993bde095",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "605f04618bbc1bd6a45d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/workflow.json b/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..d642a442fe355a5d04a43413ef15e7e72a59b439
--- /dev/null
+++ b/records/sd1/sweep/50fd72d0f9c9b9b8ce7d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 50fd72d0f9c9b9b8ce7d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0.skip_connection\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5103344bf69b391b821a/measurements.json b/records/sd1/sweep/5103344bf69b391b821a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c8cf504401a9a35f040ad1fc65988846cd73a6ad
--- /dev/null
+++ b/records/sd1/sweep/5103344bf69b391b821a/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5103344bf69b391b821a",
+ "values": {
+  "std": {
+   "value": 32.892,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01982,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 4.902,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.006461,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.021226,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.012998,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.00498,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9937,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3272,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5103344bf69b391b821a/record.json b/records/sd1/sweep/5103344bf69b391b821a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..fb7cd8f138084612448e48499d4cf4bdd3480ef1
--- /dev/null
+++ b/records/sd1/sweep/5103344bf69b391b821a/record.json
@@ -0,0 +1,79 @@
+{
+ "id": "5103344bf69b391b821a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:14:40Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.2,
+    "seed": 1234
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "medium",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "add_noise",
+   0.2,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "0206658499346b811c04"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5103344bf69b391b821a/workflow.json b/records/sd1/sweep/5103344bf69b391b821a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a6fa80e384395f0f2f51e2334ad42ecaf5770028
--- /dev/null
+++ b/records/sd1/sweep/5103344bf69b391b821a/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5103344bf69b391b821a)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.2,\n    \"seed\": 1234\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5104d142897abe5a77ce/measurements.json b/records/sd1/sweep/5104d142897abe5a77ce/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..424e96bcae07c92142c839f3940bd20b884cf510
--- /dev/null
+++ b/records/sd1/sweep/5104d142897abe5a77ce/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5104d142897abe5a77ce",
+ "values": {
+  "std": {
+   "value": 26.81,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00398,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 44.633,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.663293,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.751354,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.944019,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.455131,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.5848,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4887,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5104d142897abe5a77ce/record.json b/records/sd1/sweep/5104d142897abe5a77ce/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..34b753b0b69ba7246d03a0d345c4ff64c6c7adbc
--- /dev/null
+++ b/records/sd1/sweep/5104d142897abe5a77ce/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "5104d142897abe5a77ce",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:30:16Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "2976534844fe80be26b4",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "multiply",
+   0.0,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for out.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a923728ae4d6d42729ac"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5104d142897abe5a77ce/workflow.json b/records/sd1/sweep/5104d142897abe5a77ce/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..1355ebeb58b673b4d056229996b84c6fa81606d4
--- /dev/null
+++ b/records/sd1/sweep/5104d142897abe5a77ce/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5104d142897abe5a77ce)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5105cecc0c3d6ccf8f08/measurements.json b/records/sd1/sweep/5105cecc0c3d6ccf8f08/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..de80c2e0d0e5dfb24ae8b7293dcad86c0992daf5
--- /dev/null
+++ b/records/sd1/sweep/5105cecc0c3d6ccf8f08/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5105cecc0c3d6ccf8f08",
+ "values": {
+  "cosine_distance": {
+   "value": 0.309284,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.329107,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.275919,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.197491,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 63.967,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0056,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 40.814,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0326,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.328,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5105cecc0c3d6ccf8f08/record.json b/records/sd1/sweep/5105cecc0c3d6ccf8f08/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3b4cde7b07dc373a1a8c7efa8865079ba2fe8377
--- /dev/null
+++ b/records/sd1/sweep/5105cecc0c3d6ccf8f08/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "5105cecc0c3d6ccf8f08",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:33Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0.in_layers.1",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.5
+   },
+   "group": "in.hi",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/e95089523676889442273c36311f378aaf70550bba8b9078b5a1314bb4e5ab05.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "e95089523676889442273c36311f378aaf70550bba8b9078b5a1314bb4e5ab05",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "492b8d81d6d6e92f5465"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5105cecc0c3d6ccf8f08/workflow.json b/records/sd1/sweep/5105cecc0c3d6ccf8f08/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c851e397a7fb99b86b5ccf622422e7e60ec36b5a
--- /dev/null
+++ b/records/sd1/sweep/5105cecc0c3d6ccf8f08/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5105cecc0c3d6ccf8f08)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0.in_layers.1\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51081d719b54fdfa5a60/measurements.json b/records/sd1/sweep/51081d719b54fdfa5a60/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6745c8be5981c3b1b441b44f123cf28a1a9a3aca
--- /dev/null
+++ b/records/sd1/sweep/51081d719b54fdfa5a60/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "51081d719b54fdfa5a60",
+ "values": {
+  "cosine_distance": {
+   "value": 0.325001,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.338287,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.151576,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.064787,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.58,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00498,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 36.017,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9279,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0663,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51081d719b54fdfa5a60/record.json b/records/sd1/sweep/51081d719b54fdfa5a60/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a4ebeb06cb43819084e2918b707d6c2cda58e414
--- /dev/null
+++ b/records/sd1/sweep/51081d719b54fdfa5a60/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "51081d719b54fdfa5a60",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:42:12Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.7.1.transformer_blocks.0.attn2.to_out.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.lo",
+   "kind": "cross_attn",
+   "module_type": "Dropout",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/1783552ce22886441ef90a1b7ee63b707fdc96460b557b59a13448e074b18e77.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "1783552ce22886441ef90a1b7ee63b707fdc96460b557b59a13448e074b18e77",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "18dcf7d38983fbac8599"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51081d719b54fdfa5a60/workflow.json b/records/sd1/sweep/51081d719b54fdfa5a60/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..5b4b72d1c27331a8adc59b14aab01a5fb9e29f2c
--- /dev/null
+++ b/records/sd1/sweep/51081d719b54fdfa5a60/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51081d719b54fdfa5a60)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.7.1.transformer_blocks.0.attn2.to_out.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5109fd148368c3c0b021/measurements.json b/records/sd1/sweep/5109fd148368c3c0b021/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..9e5d9181cfc61921f8b7e6b965a5dff0e51eee6b
--- /dev/null
+++ b/records/sd1/sweep/5109fd148368c3c0b021/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5109fd148368c3c0b021",
+ "values": {
+  "std": {
+   "value": 48.296,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01249,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 15.217,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.173434,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.278668,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.259475,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.039948,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0067,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.024,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5109fd148368c3c0b021/record.json b/records/sd1/sweep/5109fd148368c3c0b021/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b4b5c0cca6f68e75034ca3490d7ed571903e6f
--- /dev/null
+++ b/records/sd1/sweep/5109fd148368c3c0b021/record.json
@@ -0,0 +1,77 @@
+{
+ "id": "5109fd148368c3c0b021",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:38:47Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "mid",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "add_scalar",
+   0.8,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5109fd148368c3c0b021/workflow.json b/records/sd1/sweep/5109fd148368c3c0b021/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..2b43e9c3d7154915d872b6b613ae801e8db209b7
--- /dev/null
+++ b/records/sd1/sweep/5109fd148368c3c0b021/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5109fd148368c3c0b021)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5110c17db168326a43bd/interpretations.jsonl b/records/sd1/sweep/5110c17db168326a43bd/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..8cd3abb7e190eaf2bdc6cf569e4899a5d3fdec1e
--- /dev/null
+++ b/records/sd1/sweep/5110c17db168326a43bd/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "5110c17db168326a43bd"}, "kind": "caption", "value": "Faded, ghostly sepia portrait with blurred face, wild hair and stained aged plate edges", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:15:46Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "e72d19af584a28ed"}
+{"target": {"record": "5110c17db168326a43bd"}, "kind": "change", "value": "The portrait turns into a washed-out, decayed early photograph, with a soft, smeared face, frizzy hair, a brownish tint and stained edges.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:15:46Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3354a76689dc9e0f"}
+{"target": {"record": "5110c17db168326a43bd"}, "kind": "keywords", "value": ["ghostly portrait", "faded photograph", "decayed plate", "soft blur", "antique"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:15:46Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "5b941a45af54a230"}
+{"target": {"record": "5110c17db168326a43bd"}, "kind": "effect_tags", "value": ["washed_out", "blur", "texture", "desaturated", "style_change", "distorted"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:15:46Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "07208effa3aa485b"}
diff --git a/records/sd1/sweep/5110c17db168326a43bd/measurements.json b/records/sd1/sweep/5110c17db168326a43bd/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..98aaf89e5fa079e6cb28c9c77c0398183e4964f6
--- /dev/null
+++ b/records/sd1/sweep/5110c17db168326a43bd/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5110c17db168326a43bd",
+ "values": {
+  "cosine_distance": {
+   "value": 0.310578,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.331873,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.235082,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.146106,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 43.684,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00325,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 46.867,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0012,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1125,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5110c17db168326a43bd/record.json b/records/sd1/sweep/5110c17db168326a43bd/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..016f343d1e238a36d0eeee35f1a331f953c6f43d
--- /dev/null
+++ b/records/sd1/sweep/5110c17db168326a43bd/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "5110c17db168326a43bd",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:54Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1.norm",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "out.hi",
+   "kind": "norm",
+   "module_type": "GroupNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/4b39216a63a59d67f3cbcb1b149833c93acec61075583ec6d260181e6a8e9dfe.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "4b39216a63a59d67f3cbcb1b149833c93acec61075583ec6d260181e6a8e9dfe",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "16709fc6fc5f8270c896"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5110c17db168326a43bd/workflow.json b/records/sd1/sweep/5110c17db168326a43bd/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a690186b6bd7c2c3fb74dfb90e8247be1d069bcc
--- /dev/null
+++ b/records/sd1/sweep/5110c17db168326a43bd/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5110c17db168326a43bd)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1.norm\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/511edb7d246d4e38c4e0/measurements.json b/records/sd1/sweep/511edb7d246d4e38c4e0/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..bb0c7db3f82a2bf266259d632bec2d15214b18e0
--- /dev/null
+++ b/records/sd1/sweep/511edb7d246d4e38c4e0/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "511edb7d246d4e38c4e0",
+ "values": {
+  "cosine_distance": {
+   "value": 0.641172,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.650734,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.921518,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.434363,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 47.275,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01661,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 58.695,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.6544,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5232,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/511edb7d246d4e38c4e0/record.json b/records/sd1/sweep/511edb7d246d4e38c4e0/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..618ef4ceab1e4d1f118a27ba2a86674c2421dc51
--- /dev/null
+++ b/records/sd1/sweep/511edb7d246d4e38c4e0/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "511edb7d246d4e38c4e0",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:30Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.5.2.conv",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "out.lo",
+   "kind": "upsample",
+   "module_type": "Conv2d",
+   "container": "Upsample",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/cd04cb1dc4aca5cf71bf6faae5f19098ed9a53981561d4ce548d30001d326b17.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "cd04cb1dc4aca5cf71bf6faae5f19098ed9a53981561d4ce548d30001d326b17",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.Upsample.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "3be815ffde4b28eda8b3"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/511edb7d246d4e38c4e0/workflow.json b/records/sd1/sweep/511edb7d246d4e38c4e0/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..86094150eb670732906123d0d2fa65cc0fff8570
--- /dev/null
+++ b/records/sd1/sweep/511edb7d246d4e38c4e0/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 511edb7d246d4e38c4e0)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.5.2.conv\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51240214de9c1db81359/measurements.json b/records/sd1/sweep/51240214de9c1db81359/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..48eb096c80f73f37bdc82f92d6a3f49a9c13bb4d
--- /dev/null
+++ b/records/sd1/sweep/51240214de9c1db81359/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "51240214de9c1db81359",
+ "values": {
+  "cosine_distance": {
+   "value": 0.198647,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.545247,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.23893,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.15028,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 46.306,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0015,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 23.098,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0001,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.8195,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51240214de9c1db81359/record.json b/records/sd1/sweep/51240214de9c1db81359/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1158cb2fe2a86d2032810a65e169cace9cfaf1f6
--- /dev/null
+++ b/records/sd1/sweep/51240214de9c1db81359/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "51240214de9c1db81359",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:51:00Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.11.0.in_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/f12bebae343149d70338f9354024743d164ce5829df7d566f9f70fde821159ac.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "f12bebae343149d70338f9354024743d164ce5829df7d566f9f70fde821159ac",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "ec5c00c8e3e88bbb17d9"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51240214de9c1db81359/workflow.json b/records/sd1/sweep/51240214de9c1db81359/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..06b62df54d33d40c36976b390e1c12e40e7da9c2
--- /dev/null
+++ b/records/sd1/sweep/51240214de9c1db81359/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51240214de9c1db81359)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.11.0.in_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/512663e731992e7ac9da/measurements.json b/records/sd1/sweep/512663e731992e7ac9da/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..3ad94e6725e8205368d33da5a1c2f6fe3dea1529
--- /dev/null
+++ b/records/sd1/sweep/512663e731992e7ac9da/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "512663e731992e7ac9da",
+ "values": {
+  "cosine_distance": {
+   "value": 0.40525,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.418931,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.360151,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.256787,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 63.201,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00533,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 40.707,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9585,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0372,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/512663e731992e7ac9da/record.json b/records/sd1/sweep/512663e731992e7ac9da/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d3269b44823bf3a62b0c4db26f07235eb51abf36
--- /dev/null
+++ b/records/sd1/sweep/512663e731992e7ac9da/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "512663e731992e7ac9da",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.7.1.norm",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.5
+   },
+   "group": "in.lo",
+   "kind": "norm",
+   "module_type": "GroupNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/dc00b55cde8a04c822e9f51b91863940c7e36194a3632c194f81cdc515e1d262.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "dc00b55cde8a04c822e9f51b91863940c7e36194a3632c194f81cdc515e1d262",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "e169994cd3b097b2cd89"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/512663e731992e7ac9da/workflow.json b/records/sd1/sweep/512663e731992e7ac9da/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..eb290146e917b455bfb5896f89817e2054382f80
--- /dev/null
+++ b/records/sd1/sweep/512663e731992e7ac9da/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 512663e731992e7ac9da)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.7.1.norm\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/513fc3a8dfe2f1e035d1/measurements.json b/records/sd1/sweep/513fc3a8dfe2f1e035d1/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..be33a5302fe6f0327a65094202828bdda5ce7566
--- /dev/null
+++ b/records/sd1/sweep/513fc3a8dfe2f1e035d1/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "513fc3a8dfe2f1e035d1",
+ "values": {
+  "std": {
+   "value": 44.442,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01701,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 4.378,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.011097,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.042883,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.009682,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.003687,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9926,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0675,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/513fc3a8dfe2f1e035d1/record.json b/records/sd1/sweep/513fc3a8dfe2f1e035d1/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..201322de54d79b4f4c0703eb68ee5f8b8e8964c5
--- /dev/null
+++ b/records/sd1/sweep/513fc3a8dfe2f1e035d1/record.json
@@ -0,0 +1,76 @@
+{
+ "id": "513fc3a8dfe2f1e035d1",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:54:32Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "mid"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "add_scalar",
+   0.8,
+   "all"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "42f1f976c2b9f23630bc"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/513fc3a8dfe2f1e035d1/workflow.json b/records/sd1/sweep/513fc3a8dfe2f1e035d1/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ff2a483fb1040b85f43b0d67c67934a8b6008da0
--- /dev/null
+++ b/records/sd1/sweep/513fc3a8dfe2f1e035d1/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 513fc3a8dfe2f1e035d1)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5140f05cbee25238e941/interpretations.jsonl b/records/sd1/sweep/5140f05cbee25238e941/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..29f0ee9ff8e90aa064b8f35823031cdcd828d3ca
--- /dev/null
+++ b/records/sd1/sweep/5140f05cbee25238e941/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "5140f05cbee25238e941"}, "kind": "caption", "value": "Smeared, swirling monochrome face with teal and cream streaks on a canvas-like textured ground", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:13:19Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "92425186b1024c40"}
+{"target": {"record": "5140f05cbee25238e941"}, "kind": "change", "value": "The crisp studio portrait dissolves into a tilted, swirling smear where the face is only barely readable, and the palette moves to cream, black and teal with a woven canvas texture.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:13:19Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "f65dd573cbec1f22"}
+{"target": {"record": "5140f05cbee25238e941"}, "kind": "keywords", "value": ["swirled face", "smeared portrait", "teal cream", "canvas texture", "distorted portrait"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:13:19Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "289fda99e9aef585"}
+{"target": {"record": "5140f05cbee25238e941"}, "kind": "effect_tags", "value": ["distorted", "painterly", "texture", "colour_shift", "new_composition", "abstract"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:13:19Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "54a152cd558e9806"}
diff --git a/records/sd1/sweep/5140f05cbee25238e941/measurements.json b/records/sd1/sweep/5140f05cbee25238e941/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..555f062cde8642867b66307168471d1714532b78
--- /dev/null
+++ b/records/sd1/sweep/5140f05cbee25238e941/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5140f05cbee25238e941",
+ "values": {
+  "cosine_distance": {
+   "value": 0.611012,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.720234,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.651234,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.333906,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 62.065,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00399,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 60.502,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.94,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.554,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5140f05cbee25238e941/record.json b/records/sd1/sweep/5140f05cbee25238e941/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..789449764a3798012b6e99c26ebf2945486d9d75
--- /dev/null
+++ b/records/sd1/sweep/5140f05cbee25238e941/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "5140f05cbee25238e941",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:16Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.4.0.skip_connection",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "in.mid",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/f0400a778aa016d3e4b84673f93c3cadd0bd88bc7bb0f3fe49457f2a0161af6d.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "f0400a778aa016d3e4b84673f93c3cadd0bd88bc7bb0f3fe49457f2a0161af6d",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "d93cdf5fd0ce8e1ad01a"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5140f05cbee25238e941/workflow.json b/records/sd1/sweep/5140f05cbee25238e941/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..834228c02d09b757b825eb2784c6145d7ddb6ce9
--- /dev/null
+++ b/records/sd1/sweep/5140f05cbee25238e941/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5140f05cbee25238e941)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.4.0.skip_connection\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5144820997dd49afa9a9/measurements.json b/records/sd1/sweep/5144820997dd49afa9a9/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..42f1a3d1dfdfbd9bb455864b0de673667ce3ae0b
--- /dev/null
+++ b/records/sd1/sweep/5144820997dd49afa9a9/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5144820997dd49afa9a9",
+ "values": {
+  "cosine_distance": {
+   "value": 0.227793,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.302153,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.157236,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.047428,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 56.24,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00933,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 26.944,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9094,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3022,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5144820997dd49afa9a9/record.json b/records/sd1/sweep/5144820997dd49afa9a9/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a4237285fdc7309ca3b88f9d1fd1115564d1909f
--- /dev/null
+++ b/records/sd1/sweep/5144820997dd49afa9a9/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5144820997dd49afa9a9",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:32Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.1.transformer_blocks.0.norm1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "out.mid",
+   "kind": "norm",
+   "module_type": "LayerNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/74c8f591ec9fb8f5d8936c2fee180057a1f2df7882e0039daa8160947ff1d5f6.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "74c8f591ec9fb8f5d8936c2fee180057a1f2df7882e0039daa8160947ff1d5f6",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.LayerNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "3b235644ec419cac68e7"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5144820997dd49afa9a9/workflow.json b/records/sd1/sweep/5144820997dd49afa9a9/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..16211d1bf65171bd35d107fe8d9ad7f7efe643b7
--- /dev/null
+++ b/records/sd1/sweep/5144820997dd49afa9a9/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5144820997dd49afa9a9)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.1.transformer_blocks.0.norm1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/514eb1040f7efe573438/measurements.json b/records/sd1/sweep/514eb1040f7efe573438/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ea915cba5d605b6de683a2d23ed08d664343435
--- /dev/null
+++ b/records/sd1/sweep/514eb1040f7efe573438/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "514eb1040f7efe573438",
+ "values": {
+  "cosine_distance": {
+   "value": 0.246672,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.280139,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.123361,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.065633,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 62.982,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00314,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.089,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0116,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2848,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/514eb1040f7efe573438/record.json b/records/sd1/sweep/514eb1040f7efe573438/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..aed685cb11fb016145ca5fe87b5e6d79967c7fcf
--- /dev/null
+++ b/records/sd1/sweep/514eb1040f7efe573438/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "514eb1040f7efe573438",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:33Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.10.0.out_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/80dbe80c3c225b170f4855d7f05f21908ca65409367d0cccc1de02838d8c0bc3.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "80dbe80c3c225b170f4855d7f05f21908ca65409367d0cccc1de02838d8c0bc3",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "21fadcfe63ac471c70d5"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/514eb1040f7efe573438/workflow.json b/records/sd1/sweep/514eb1040f7efe573438/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..13e7ddf26c52b2dbe98f22c59dcdfe228d1666ec
--- /dev/null
+++ b/records/sd1/sweep/514eb1040f7efe573438/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 514eb1040f7efe573438)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.10.0.out_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5158af0cc941efd46c62/measurements.json b/records/sd1/sweep/5158af0cc941efd46c62/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..40de3514852ae2134d8152d260f4c4af5d5861d4
--- /dev/null
+++ b/records/sd1/sweep/5158af0cc941efd46c62/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5158af0cc941efd46c62",
+ "values": {
+  "cosine_distance": {
+   "value": 0.002929,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.00642,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.005414,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.001492,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.023,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00561,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 2.203,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0127,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1231,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5158af0cc941efd46c62/record.json b/records/sd1/sweep/5158af0cc941efd46c62/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ead68cd09703973fd6d011c79cdb526f3c8b0f5e
--- /dev/null
+++ b/records/sd1/sweep/5158af0cc941efd46c62/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5158af0cc941efd46c62",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:28Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.1.transformer_blocks.0.attn2.to_k",
+   "op": "multiply",
+   "args": {
+    "scalar": 1.5
+   },
+   "group": "out.mid",
+   "kind": "cross_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/5e0c2354c7f6baf02497305627321971c97adb870de1a2b3cdb9886a329f987b.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "5e0c2354c7f6baf02497305627321971c97adb870de1a2b3cdb9886a329f987b",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "5ecb48e428a21fe641fe"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5158af0cc941efd46c62/workflow.json b/records/sd1/sweep/5158af0cc941efd46c62/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ef7e260bf4cede49bf086665227f6ddd95dc0d20
--- /dev/null
+++ b/records/sd1/sweep/5158af0cc941efd46c62/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5158af0cc941efd46c62)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.1.transformer_blocks.0.attn2.to_k\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/515c4272c6f1df2d9b0a/measurements.json b/records/sd1/sweep/515c4272c6f1df2d9b0a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..5fdf82c2ac41e0a0e3815fc2b779cc076e3bfd2a
--- /dev/null
+++ b/records/sd1/sweep/515c4272c6f1df2d9b0a/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "515c4272c6f1df2d9b0a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.433073,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.503072,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.354754,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.255395,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 59.303,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00467,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 42.372,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8816,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.511,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/515c4272c6f1df2d9b0a/record.json b/records/sd1/sweep/515c4272c6f1df2d9b0a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..98afb759a5684ff67db644a04f3d4e1f126f1624
--- /dev/null
+++ b/records/sd1/sweep/515c4272c6f1df2d9b0a/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "515c4272c6f1df2d9b0a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:48:48Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.7.1.norm",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "in.lo",
+   "kind": "norm",
+   "module_type": "GroupNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/70c8d6919de208901964823804072617221cba4d4af084b281f2de930f59aad1.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "70c8d6919de208901964823804072617221cba4d4af084b281f2de930f59aad1",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "00fbbd69d2e2dcea60be"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/515c4272c6f1df2d9b0a/workflow.json b/records/sd1/sweep/515c4272c6f1df2d9b0a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5011c980e92de56bca8180110b8d8bd11c272bb
--- /dev/null
+++ b/records/sd1/sweep/515c4272c6f1df2d9b0a/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 515c4272c6f1df2d9b0a)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.7.1.norm\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/516cf415983aac9760dc/measurements.json b/records/sd1/sweep/516cf415983aac9760dc/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..28ddddcc6de4e46f681ff432f50d65791a596067
--- /dev/null
+++ b/records/sd1/sweep/516cf415983aac9760dc/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "516cf415983aac9760dc",
+ "values": {
+  "cosine_distance": {
+   "value": 0.118194,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.152808,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.075586,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.040298,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 70.071,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00586,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 19.7,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.034,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0762,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/516cf415983aac9760dc/record.json b/records/sd1/sweep/516cf415983aac9760dc/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..225a6c62704d6ce526247143ad44f85857c24f57
--- /dev/null
+++ b/records/sd1/sweep/516cf415983aac9760dc/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "516cf415983aac9760dc",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:47:08Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1.transformer_blocks.0.attn1.to_out.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.hi",
+   "kind": "self_attn",
+   "module_type": "Dropout",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/3522376b6b4fa58131ef0991545f1055376ac5d63d0ddacad8a7ea68ea36621a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "3522376b6b4fa58131ef0991545f1055376ac5d63d0ddacad8a7ea68ea36621a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "ef2d532eb673f9527906"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/516cf415983aac9760dc/workflow.json b/records/sd1/sweep/516cf415983aac9760dc/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7e69eb9e24252e30b4dc0ec48fd6a052c5ee6263
--- /dev/null
+++ b/records/sd1/sweep/516cf415983aac9760dc/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 516cf415983aac9760dc)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1.transformer_blocks.0.attn1.to_out.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/516d1f6b971d3307b227/measurements.json b/records/sd1/sweep/516d1f6b971d3307b227/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..8c01f6b04c793326b94fa263208d21fcd78cca07
--- /dev/null
+++ b/records/sd1/sweep/516d1f6b971d3307b227/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "516d1f6b971d3307b227",
+ "values": {
+  "cosine_distance": {
+   "value": 0.123801,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.111137,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.067434,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.051564,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 64.155,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0055,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 13.204,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0198,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3889,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/516d1f6b971d3307b227/record.json b/records/sd1/sweep/516d1f6b971d3307b227/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1b5191959c971a07da29ff5c99d4b19afeea9d7d
--- /dev/null
+++ b/records/sd1/sweep/516d1f6b971d3307b227/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "516d1f6b971d3307b227",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:37Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "middle_block.0.in_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/81458457e4e6294913e7fbd41a9c7cfabf70e9221260fa2349ce361049327961.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "81458457e4e6294913e7fbd41a9c7cfabf70e9221260fa2349ce361049327961",
+  "type_path": "UNetModel.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "3f7a8d7a4b66cbc5ea3e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/516d1f6b971d3307b227/workflow.json b/records/sd1/sweep/516d1f6b971d3307b227/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..2f3e1d068f6e36ada7b75a01fd8d5c54d5a1c402
--- /dev/null
+++ b/records/sd1/sweep/516d1f6b971d3307b227/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 516d1f6b971d3307b227)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.0.in_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5172bab501a82cf53045/measurements.json b/records/sd1/sweep/5172bab501a82cf53045/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2588827bde119d42c25142666f717d49e35d4f2c
--- /dev/null
+++ b/records/sd1/sweep/5172bab501a82cf53045/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5172bab501a82cf53045",
+ "values": {
+  "std": {
+   "value": 43.937,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0166,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 7.378,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.033508,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.07755,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.020116,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.015903,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0103,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0148,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5172bab501a82cf53045/record.json b/records/sd1/sweep/5172bab501a82cf53045/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7088c0abb740b7fa9aba5c9fb1825e58d3df945e
--- /dev/null
+++ b/records/sd1/sweep/5172bab501a82cf53045/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "5172bab501a82cf53045",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:56:34Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.1",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "out.lo",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "mid",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.lo",
+   "add_scalar",
+   0.8,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "98b86dcbd5aad0aaa6c5"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5172bab501a82cf53045/workflow.json b/records/sd1/sweep/5172bab501a82cf53045/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..63afb085868449d546db3ae395172b6702b3c12e
--- /dev/null
+++ b/records/sd1/sweep/5172bab501a82cf53045/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5172bab501a82cf53045)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.1\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51773c061de31c442b88/measurements.json b/records/sd1/sweep/51773c061de31c442b88/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..17e125f4b3995b8114500068ceec596ed5a6d1ec
--- /dev/null
+++ b/records/sd1/sweep/51773c061de31c442b88/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "51773c061de31c442b88",
+ "values": {
+  "cosine_distance": {
+   "value": 0.54415,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.572539,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.667698,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.363051,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 28.754,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00304,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 45.853,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9713,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7891,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51773c061de31c442b88/record.json b/records/sd1/sweep/51773c061de31c442b88/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..f11c766e8dcec271084a98c19aacb8a5374189b7
--- /dev/null
+++ b/records/sd1/sweep/51773c061de31c442b88/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "51773c061de31c442b88",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:22Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.4.1.transformer_blocks.0.attn1.to_q",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "out.lo",
+   "kind": "self_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/553ad9b30221a10c0c1afb6e5b9505d0da7c047996b9303fae508d4f6942ef67.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "553ad9b30221a10c0c1afb6e5b9505d0da7c047996b9303fae508d4f6942ef67",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "315c489ea92e75dc50b3"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51773c061de31c442b88/workflow.json b/records/sd1/sweep/51773c061de31c442b88/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..aca9b9edf17af632aa02d5c09011498539c4f60e
--- /dev/null
+++ b/records/sd1/sweep/51773c061de31c442b88/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51773c061de31c442b88)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.4.1.transformer_blocks.0.attn1.to_q\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51776f4c8339d1fe1963/interpretations.jsonl b/records/sd1/sweep/51776f4c8339d1fe1963/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..dbbe72f9099b2238bfd04cd96aa2a0223a1dc9ef
--- /dev/null
+++ b/records/sd1/sweep/51776f4c8339d1fe1963/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "51776f4c8339d1fe1963"}, "kind": "caption", "value": "Short-haired woman in black shirt, grey backdrop, a little softer and lower in contrast", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:19:42Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3b3062e463c9c26d"}
+{"target": {"record": "51776f4c8339d1fe1963"}, "kind": "change", "value": "Almost no visible change; the image is slightly softer and lower in contrast, with a slightly paler background.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:19:42Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a94ea3f0cd7bae85"}
+{"target": {"record": "51776f4c8339d1fe1963"}, "kind": "keywords", "value": ["portrait", "near identical", "soft contrast", "pale backdrop"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:19:42Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "42c4fb7e293bf17d"}
+{"target": {"record": "51776f4c8339d1fe1963"}, "kind": "effect_tags", "value": ["subtle", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:19:42Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a1c037b7f34d8aef"}
diff --git a/records/sd1/sweep/51776f4c8339d1fe1963/measurements.json b/records/sd1/sweep/51776f4c8339d1fe1963/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..d6fea7dda31a7413721015b307e10c135ebff094
--- /dev/null
+++ b/records/sd1/sweep/51776f4c8339d1fe1963/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "51776f4c8339d1fe1963",
+ "values": {
+  "cosine_distance": {
+   "value": 0.0,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.0,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": -0.0,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 66.477,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00637,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.0,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0093,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2957,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51776f4c8339d1fe1963/record.json b/records/sd1/sweep/51776f4c8339d1fe1963/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3df77a9abf7a9121082370af51340c321bcade8d
--- /dev/null
+++ b/records/sd1/sweep/51776f4c8339d1fe1963/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "51776f4c8339d1fe1963",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:03Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.2.0.h_upd",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.5
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/1f6506b352f28887e13446cc285f0d800b52ba1acb81e2b4a3226fb0a96bab3c.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "1f6506b352f28887e13446cc285f0d800b52ba1acb81e2b4a3226fb0a96bab3c",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "15e98ab258aef9c58ee1"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51776f4c8339d1fe1963/workflow.json b/records/sd1/sweep/51776f4c8339d1fe1963/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..6b393f7923c326f3f73560ed229b32dd1fd089f8
--- /dev/null
+++ b/records/sd1/sweep/51776f4c8339d1fe1963/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51776f4c8339d1fe1963)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.2.0.h_upd\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5184e8708316ffec9ed5/measurements.json b/records/sd1/sweep/5184e8708316ffec9ed5/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..da76f9a8a90bcac70a0073ae75422ef5bd21d2ff
--- /dev/null
+++ b/records/sd1/sweep/5184e8708316ffec9ed5/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5184e8708316ffec9ed5",
+ "values": {
+  "std": {
+   "value": 37.617,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01365,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 5.747,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.021005,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.058486,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.013322,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.004547,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.998,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0134,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5184e8708316ffec9ed5/record.json b/records/sd1/sweep/5184e8708316ffec9ed5/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7dc914145228b2c5f28afcd222532b2b4f529943
--- /dev/null
+++ b/records/sd1/sweep/5184e8708316ffec9ed5/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "5184e8708316ffec9ed5",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:02:00Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.2,
+    "seed": 1234
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "medium",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "add_noise",
+   0.2,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5184e8708316ffec9ed5/workflow.json b/records/sd1/sweep/5184e8708316ffec9ed5/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..27f3934f83de5344e6b0ab2c7dae69c313dd0f7f
--- /dev/null
+++ b/records/sd1/sweep/5184e8708316ffec9ed5/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5184e8708316ffec9ed5)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.2,\n    \"seed\": 1234\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51850edc8447eed7c6c8/interpretations.jsonl b/records/sd1/sweep/51850edc8447eed7c6c8/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..08dedaf619c49eaa1cc4e36e0f2a542b5b59c8d0
--- /dev/null
+++ b/records/sd1/sweep/51850edc8447eed7c6c8/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "51850edc8447eed7c6c8"}, "kind": "caption", "value": "Blurry abstract painting of orange-brown vertical form against blue and teal smears", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:05:27Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "778ad46347ec52cf"}
+{"target": {"record": "51850edc8447eed7c6c8"}, "kind": "change", "value": "The portrait dissolves into a soft, blurred abstract of warm orange-brown strokes running up the centre, surrounded by cool blue-green smudges.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:05:27Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ca59eca9e81c60ce"}
+{"target": {"record": "51850edc8447eed7c6c8"}, "kind": "keywords", "value": ["blurred abstraction", "warm cool contrast", "smeared paint", "portrait lost", "soft focus"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:05:27Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "6c56eb5d1f8c0c3f"}
+{"target": {"record": "51850edc8447eed7c6c8"}, "kind": "effect_tags", "value": ["abstract", "blur", "painterly", "subject_lost", "colour_shift"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:05:27Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "4b15abc3d385590f"}
diff --git a/records/sd1/sweep/51850edc8447eed7c6c8/measurements.json b/records/sd1/sweep/51850edc8447eed7c6c8/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..42ca31a7f5f1387efe508f8b7889bd471481c696
--- /dev/null
+++ b/records/sd1/sweep/51850edc8447eed7c6c8/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "51850edc8447eed7c6c8",
+ "values": {
+  "cosine_distance": {
+   "value": 0.624677,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.762659,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.83056,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.426493,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.323,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0033,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 57.472,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.7749,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4785,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51850edc8447eed7c6c8/record.json b/records/sd1/sweep/51850edc8447eed7c6c8/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9daae0155f5feaa9e95d6d40ae0cc99fbbaf3c41
--- /dev/null
+++ b/records/sd1/sweep/51850edc8447eed7c6c8/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "51850edc8447eed7c6c8",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:36Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.3.0.op",
+   "op": "scale",
+   "args": {
+    "scale_factor": 2
+   },
+   "group": "in.hi",
+   "kind": "downsample",
+   "module_type": "Conv2d",
+   "container": "Downsample",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "strong_grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/6de0734656cf758e61990bb716181d5d8d1bd975d18193d51156098196923f11.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "6de0734656cf758e61990bb716181d5d8d1bd975d18193d51156098196923f11",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.Downsample.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "2e10951c3d74693a5055"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51850edc8447eed7c6c8/workflow.json b/records/sd1/sweep/51850edc8447eed7c6c8/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..fea06944ecfc601fdd5bbb800331356846498edb
--- /dev/null
+++ b/records/sd1/sweep/51850edc8447eed7c6c8/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51850edc8447eed7c6c8)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.3.0.op\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5196b0938c7f4b90c5c0/interpretations.jsonl b/records/sd1/sweep/5196b0938c7f4b90c5c0/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..107bef0f403bfc1b6841e4bf2162a03c65d5ca32
--- /dev/null
+++ b/records/sd1/sweep/5196b0938c7f4b90c5c0/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "5196b0938c7f4b90c5c0"}, "kind": "caption", "value": "Blurry, tiled mosaic face in muddy browns with blue flecks and a dark curly hair mass", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:25:26Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "898e37d6d1461b75"}
+{"target": {"record": "5196b0938c7f4b90c5c0"}, "kind": "change", "value": "The crisp illustration dissolves into a murky, blurred mosaic; the face survives only as a ghostly impression.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:25:26Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "caed04c18446f7ff"}
+{"target": {"record": "5196b0938c7f4b90c5c0"}, "kind": "keywords", "value": ["mosaic", "ghost face", "murky", "tiled blur", "dissolved portrait"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:25:26Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "5dfe687860426cc0"}
+{"target": {"record": "5196b0938c7f4b90c5c0"}, "kind": "effect_tags", "value": ["blur", "fragmented", "texture", "abstract", "distorted", "colour_shift"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:25:26Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "57c72771f6422477"}
diff --git a/records/sd1/sweep/5196b0938c7f4b90c5c0/measurements.json b/records/sd1/sweep/5196b0938c7f4b90c5c0/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c6dc5175099f8984e105a5f02678e0d4aff80f1
--- /dev/null
+++ b/records/sd1/sweep/5196b0938c7f4b90c5c0/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5196b0938c7f4b90c5c0",
+ "values": {
+  "cosine_distance": {
+   "value": 0.551861,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.559846,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.565604,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.227356,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 31.773,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00473,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 40.908,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0321,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4234,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5196b0938c7f4b90c5c0/record.json b/records/sd1/sweep/5196b0938c7f4b90c5c0/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..489d84488dc4455f5584fb528342377409115dcc
--- /dev/null
+++ b/records/sd1/sweep/5196b0938c7f4b90c5c0/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5196b0938c7f4b90c5c0",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:47:40Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "out.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.conv",
+   "kind": "out_conv",
+   "module_type": "SiLU",
+   "container": "Sequential",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/159df83cf744814696bf263a274ee3968acc7c51e1374706cdcbe72f0b94b5aa.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "159df83cf744814696bf263a274ee3968acc7c51e1374706cdcbe72f0b94b5aa",
+  "type_path": "UNetModel.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "154e080e9bcfdb9d2035"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5196b0938c7f4b90c5c0/workflow.json b/records/sd1/sweep/5196b0938c7f4b90c5c0/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5830cad7d6746ffb5f6ececfabe5ebc457b0bf6
--- /dev/null
+++ b/records/sd1/sweep/5196b0938c7f4b90c5c0/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5196b0938c7f4b90c5c0)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"out.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51a212e06965e124d9c1/measurements.json b/records/sd1/sweep/51a212e06965e124d9c1/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f87103251da3fae42c20995ad36fe3abb76c33bc
--- /dev/null
+++ b/records/sd1/sweep/51a212e06965e124d9c1/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "51a212e06965e124d9c1",
+ "values": {
+  "std": {
+   "value": 37.415,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00585,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 44.387,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.576949,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.689208,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.265118,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.145478,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9925,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5111,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51a212e06965e124d9c1/record.json b/records/sd1/sweep/51a212e06965e124d9c1/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9a98b2f17b4e4de21392613eb8d758b73ab6082c
--- /dev/null
+++ b/records/sd1/sweep/51a212e06965e124d9c1/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "51a212e06965e124d9c1",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:34:30Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": -0.8
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "mid"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "add_scalar",
+   -0.8,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": -0.8,
+    "applied": -0.8,
+    "reason": "outside safe range [-0.5, 1.5] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "363692deecd7384f259a"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51a212e06965e124d9c1/workflow.json b/records/sd1/sweep/51a212e06965e124d9c1/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d206ca0e5c1d11dc225773df662472ac8427716
--- /dev/null
+++ b/records/sd1/sweep/51a212e06965e124d9c1/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51a212e06965e124d9c1)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": -0.8\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51a55d105ab00a492609/measurements.json b/records/sd1/sweep/51a55d105ab00a492609/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..b1f883d9c0325fcadf57d552d34468dcb346213c
--- /dev/null
+++ b/records/sd1/sweep/51a55d105ab00a492609/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "51a55d105ab00a492609",
+ "values": {
+  "std": {
+   "value": 53.78,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01249,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 30.383,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.258835,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.293219,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.085352,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.021969,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9879,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2267,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51a55d105ab00a492609/record.json b/records/sd1/sweep/51a55d105ab00a492609/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b3227a8b98a98145085b5cfd1cca10960a9f3153
--- /dev/null
+++ b/records/sd1/sweep/51a55d105ab00a492609/record.json
@@ -0,0 +1,76 @@
+{
+ "id": "51a55d105ab00a492609",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:39:34Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2.0
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "multiply",
+   2.0,
+   "all"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "27689ce41c1241ed68ba"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51a55d105ab00a492609/workflow.json b/records/sd1/sweep/51a55d105ab00a492609/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..53663bdd982cdead0e027622a218ef43b1486f08
--- /dev/null
+++ b/records/sd1/sweep/51a55d105ab00a492609/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51a55d105ab00a492609)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2.0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51ac9743edb53470d07a/measurements.json b/records/sd1/sweep/51ac9743edb53470d07a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e30da50465f879c92218d48c2e676e12f53a1ee6
--- /dev/null
+++ b/records/sd1/sweep/51ac9743edb53470d07a/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "51ac9743edb53470d07a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.0,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.0,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": -0.0,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 66.477,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00637,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.0,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0093,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2957,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51ac9743edb53470d07a/record.json b/records/sd1/sweep/51ac9743edb53470d07a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ae9e1333795bcead5e7a62013dcdb69fc08176cc
--- /dev/null
+++ b/records/sd1/sweep/51ac9743edb53470d07a/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "51ac9743edb53470d07a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:17Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.0.h_upd",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/1dde1a72f398c83b2b16731de9a221a13b35aa9b5cb3d8c9376f92c30d32ec27.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "1dde1a72f398c83b2b16731de9a221a13b35aa9b5cb3d8c9376f92c30d32ec27",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "c417c519f6adfad86d5f"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51ac9743edb53470d07a/workflow.json b/records/sd1/sweep/51ac9743edb53470d07a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..432d5d3ccf2ee980ed7d0f588d40768dff5e1a55
--- /dev/null
+++ b/records/sd1/sweep/51ac9743edb53470d07a/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51ac9743edb53470d07a)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.0.h_upd\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51ae2c27608c3ebca31d/interpretations.jsonl b/records/sd1/sweep/51ae2c27608c3ebca31d/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..c63546215f220cfe309dfe1c44d4f0f1ed756953
--- /dev/null
+++ b/records/sd1/sweep/51ae2c27608c3ebca31d/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "51ae2c27608c3ebca31d"}, "kind": "caption", "value": "Frontal portrait of woman with lighter wavy hair wearing a black collared shirt", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:30:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3ffeb20e88a25ccd"}
+{"target": {"record": "51ae2c27608c3ebca31d"}, "kind": "change", "value": "Very similar composition with lighter golden-brown hair, a softer face and a collared shirt replacing the turtleneck.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:30:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ccde22f75cb277d1"}
+{"target": {"record": "51ae2c27608c3ebca31d"}, "kind": "keywords", "value": ["minor variation", "lighter hair", "collared shirt", "portrait"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:30:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "b502046931ea2da9"}
+{"target": {"record": "51ae2c27608c3ebca31d"}, "kind": "effect_tags", "value": ["subtle", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:30:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a0873bc10c6fe0ce"}
diff --git a/records/sd1/sweep/51ae2c27608c3ebca31d/measurements.json b/records/sd1/sweep/51ae2c27608c3ebca31d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..4678fcdc6472d79181b837dae725a7b92e1a2a7a
--- /dev/null
+++ b/records/sd1/sweep/51ae2c27608c3ebca31d/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "51ae2c27608c3ebca31d",
+ "values": {
+  "cosine_distance": {
+   "value": 0.06466,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.096522,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.062932,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.019785,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 58.429,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.005,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 11.317,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9885,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1288,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51ae2c27608c3ebca31d/record.json b/records/sd1/sweep/51ae2c27608c3ebca31d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..e709a17ebcd36afca088dd078d23b7c9124b550f
--- /dev/null
+++ b/records/sd1/sweep/51ae2c27608c3ebca31d/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "51ae2c27608c3ebca31d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:35Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.2.0.skip_connection",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.lo",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/d723fd6ac6f3e6d0e30d2f9e0193e092706b57a9edf37f85071c15815377157a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "d723fd6ac6f3e6d0e30d2f9e0193e092706b57a9edf37f85071c15815377157a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "8980a55217a608ad4e17"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51ae2c27608c3ebca31d/workflow.json b/records/sd1/sweep/51ae2c27608c3ebca31d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..45f9fb0078194ed9825b11c7420c94bfebeb3379
--- /dev/null
+++ b/records/sd1/sweep/51ae2c27608c3ebca31d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51ae2c27608c3ebca31d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.2.0.skip_connection\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51b898c14b3abc0633c5/interpretations.jsonl b/records/sd1/sweep/51b898c14b3abc0633c5/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..92bd77cc1c06e532540feef61823391000244dc6
--- /dev/null
+++ b/records/sd1/sweep/51b898c14b3abc0633c5/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "51b898c14b3abc0633c5"}, "kind": "caption", "value": "Dense rainbow speckle noise with a faint long-haired woman's face visible through it.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:49:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "06282bf89a2a7259"}
+{"target": {"record": "51b898c14b3abc0633c5"}, "kind": "change", "value": "The clean illustrated portrait is buried under saturated colour speckle, leaving only a ghostly outline of hair and face.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:49:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "db92852d59cc65c2"}
+{"target": {"record": "51b898c14b3abc0633c5"}, "kind": "keywords", "value": ["colour noise", "ghost face", "rgb speckle", "static portrait", "dissolved"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:49:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "4fa7736bb0c549b8"}
+{"target": {"record": "51b898c14b3abc0633c5"}, "kind": "effect_tags", "value": ["glitch", "texture", "saturated", "fragmented", "degenerate", "abstract"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:49:43Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "f6d0b60f083a97ba"}
diff --git a/records/sd1/sweep/51b898c14b3abc0633c5/measurements.json b/records/sd1/sweep/51b898c14b3abc0633c5/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..a6c901295e9bdce09b8b0f641435691e913273eb
--- /dev/null
+++ b/records/sd1/sweep/51b898c14b3abc0633c5/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "51b898c14b3abc0633c5",
+ "values": {
+  "cosine_distance": {
+   "value": 0.568426,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.738291,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.744763,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.36985,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 85.873,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.05284,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 77.3,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0793,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.9803,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51b898c14b3abc0633c5/record.json b/records/sd1/sweep/51b898c14b3abc0633c5/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..c033e9835eb3854b96538ef96237884e60fcc60a
--- /dev/null
+++ b/records/sd1/sweep/51b898c14b3abc0633c5/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "51b898c14b3abc0633c5",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:47:24Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.11.0.skip_connection",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "out.hi",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/7fc2c140ee7468c269353d3b5cc9bc03b4e411a10d2a50805c246f12b3ddbb12.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "7fc2c140ee7468c269353d3b5cc9bc03b4e411a10d2a50805c246f12b3ddbb12",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "9c94ba5d7d277c6d6da9"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51b898c14b3abc0633c5/workflow.json b/records/sd1/sweep/51b898c14b3abc0633c5/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..fe5e93b66e70a41d4e4e6fb928280dc027009a3d
--- /dev/null
+++ b/records/sd1/sweep/51b898c14b3abc0633c5/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51b898c14b3abc0633c5)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.11.0.skip_connection\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51bc4fba797c49ead899/interpretations.jsonl b/records/sd1/sweep/51bc4fba797c49ead899/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..b4a302dff6189a9a3c2ec64e333c16b640cca2ad
--- /dev/null
+++ b/records/sd1/sweep/51bc4fba797c49ead899/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "51bc4fba797c49ead899"}, "kind": "caption", "value": "Blurry ghostlike figure with long dark hair and smudged black eyes on pale background", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:25:34Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "72106fa069137dc0"}
+{"target": {"record": "51bc4fba797c49ead899"}, "kind": "change", "value": "The portrait becomes a soft, out-of-focus frontal figure with long black hair and hollow dark eye smudges, giving it an eerie, ghostly look.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:25:34Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "e913e1f16be487d3"}
+{"target": {"record": "51bc4fba797c49ead899"}, "kind": "keywords", "value": ["ghostly blur", "eerie portrait", "hollow eyes", "long dark hair", "soft focus"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:25:34Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "fe56b946595b7a5d"}
+{"target": {"record": "51bc4fba797c49ead899"}, "kind": "effect_tags", "value": ["blur", "surreal", "distorted", "new_composition", "abstract", "washed_out"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:25:34Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "f559f73228b931f8"}
diff --git a/records/sd1/sweep/51bc4fba797c49ead899/measurements.json b/records/sd1/sweep/51bc4fba797c49ead899/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cf383ca9706ff78147761e0f797feb0b84a27144
--- /dev/null
+++ b/records/sd1/sweep/51bc4fba797c49ead899/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "51bc4fba797c49ead899",
+ "values": {
+  "cosine_distance": {
+   "value": 0.390741,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.551727,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.650456,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.288128,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.19,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00415,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 46.061,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9432,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.849,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51bc4fba797c49ead899/record.json b/records/sd1/sweep/51bc4fba797c49ead899/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..80380923dda984cf7ae6f400fb3cf3c713c6d3bf
--- /dev/null
+++ b/records/sd1/sweep/51bc4fba797c49ead899/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "51bc4fba797c49ead899",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:48:31Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.1.transformer_blocks.0.attn1.to_k",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "in.hi",
+   "kind": "self_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/4d4e368e40a0736cf5c790de63056fd517c9d4489a955c9f6d390d7f0dbbf72a.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "4d4e368e40a0736cf5c790de63056fd517c9d4489a955c9f6d390d7f0dbbf72a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "15d721ba04523102bf50"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51bc4fba797c49ead899/workflow.json b/records/sd1/sweep/51bc4fba797c49ead899/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c376366570f00d06e0d75a3cb683224bc086c000
--- /dev/null
+++ b/records/sd1/sweep/51bc4fba797c49ead899/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51bc4fba797c49ead899)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.1.transformer_blocks.0.attn1.to_k\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51e80a4307f990b86361/measurements.json b/records/sd1/sweep/51e80a4307f990b86361/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..d9657b16e1076e229643ced9b986059e30fb19e4
--- /dev/null
+++ b/records/sd1/sweep/51e80a4307f990b86361/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "51e80a4307f990b86361",
+ "values": {
+  "cosine_distance": {
+   "value": 0.143813,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.265526,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.175445,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.130641,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 67.129,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00535,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 17.288,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9821,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0774,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51e80a4307f990b86361/record.json b/records/sd1/sweep/51e80a4307f990b86361/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..bcb256b375fb906a9d14c720310ed30829a1a5c3
--- /dev/null
+++ b/records/sd1/sweep/51e80a4307f990b86361/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "51e80a4307f990b86361",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:28Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.11.1.proj_out",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.hi",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/9d9bdac0d97b64d6447ddf3337f98a29371c0fd1d16d92cf0e2d503f0e97d3b4.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "9d9bdac0d97b64d6447ddf3337f98a29371c0fd1d16d92cf0e2d503f0e97d3b4",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "680e9a3b51511b356cae"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51e80a4307f990b86361/workflow.json b/records/sd1/sweep/51e80a4307f990b86361/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..b802388b2625ea5725ebd1149b1a69f73de345a4
--- /dev/null
+++ b/records/sd1/sweep/51e80a4307f990b86361/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51e80a4307f990b86361)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.11.1.proj_out\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51f50524375d37c91ece/measurements.json b/records/sd1/sweep/51f50524375d37c91ece/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c9d7e8156ffcb55a57d0724fe1ccd67032ec3f38
--- /dev/null
+++ b/records/sd1/sweep/51f50524375d37c91ece/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "51f50524375d37c91ece",
+ "values": {
+  "cosine_distance": {
+   "value": 0.272924,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.329537,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.15369,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.068022,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 51.448,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00556,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 31.201,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9527,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0847,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51f50524375d37c91ece/record.json b/records/sd1/sweep/51f50524375d37c91ece/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..15c08b85e8c196158370ba0bf48f2de8f974861c
--- /dev/null
+++ b/records/sd1/sweep/51f50524375d37c91ece/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "51f50524375d37c91ece",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:44:04Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.2.0.skip_connection",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "out.lo",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/84ef8b563e28c4ae636a5b86706a4ab312d415818bd378d8ad535c2ced701899.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "84ef8b563e28c4ae636a5b86706a4ab312d415818bd378d8ad535c2ced701899",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6784eb2567eab8b0996a"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51f50524375d37c91ece/workflow.json b/records/sd1/sweep/51f50524375d37c91ece/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c60c499344b98a58af52442f91bbde7ccf8c83b
--- /dev/null
+++ b/records/sd1/sweep/51f50524375d37c91ece/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51f50524375d37c91ece)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.2.0.skip_connection\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51f841b753cef52e673c/measurements.json b/records/sd1/sweep/51f841b753cef52e673c/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c67fac59159e8bbbd8002ac29c918a565d34c980
--- /dev/null
+++ b/records/sd1/sweep/51f841b753cef52e673c/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "51f841b753cef52e673c",
+ "values": {
+  "cosine_distance": {
+   "value": 0.314562,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.334077,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.227014,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.155223,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 51.942,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00579,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 35.935,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.024,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3548,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51f841b753cef52e673c/record.json b/records/sd1/sweep/51f841b753cef52e673c/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..2472ff43cfa584118d127381dc40a8e1d6374645
--- /dev/null
+++ b/records/sd1/sweep/51f841b753cef52e673c/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "51f841b753cef52e673c",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:48Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.1.0.out_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/2b7725ce4afd2bb77b67e27d3d2fc988bc545c2efbe521ab87ff44e4370529ad.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "2b7725ce4afd2bb77b67e27d3d2fc988bc545c2efbe521ab87ff44e4370529ad",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "626d01c1e5dea2312a8e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/51f841b753cef52e673c/workflow.json b/records/sd1/sweep/51f841b753cef52e673c/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f06bb02da35e9f8b4b7f37d8eba4214816fb076d
--- /dev/null
+++ b/records/sd1/sweep/51f841b753cef52e673c/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 51f841b753cef52e673c)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.1.0.out_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5206234696cfb37d2d0d/interpretations.jsonl b/records/sd1/sweep/5206234696cfb37d2d0d/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e2e9b7a2f083077a6c8b14acd70ea7209ada8741
--- /dev/null
+++ b/records/sd1/sweep/5206234696cfb37d2d0d/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "5206234696cfb37d2d0d"}, "kind": "caption", "value": "Blurry, blotchy ghost of a woman's face framed by dark hair on beige ground", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:16:16Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "565028d862f41594"}
+{"target": {"record": "5206234696cfb37d2d0d"}, "kind": "change", "value": "The portrait melts into soft, blocky blotches on a pale beige field. Only a faint face and a dark ring of hair remain readable.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:16:16Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "33ffc7c9846a5ed0"}
+{"target": {"record": "5206234696cfb37d2d0d"}, "kind": "keywords", "value": ["blotchy blur", "ghost face", "beige", "degraded portrait", "smudged"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:16:16Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "41d3443098b34dcb"}
+{"target": {"record": "5206234696cfb37d2d0d"}, "kind": "effect_tags", "value": ["blur", "washed_out", "distorted", "degenerate", "abstract"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:16:16Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "089c3572fa7ad13a"}
diff --git a/records/sd1/sweep/5206234696cfb37d2d0d/measurements.json b/records/sd1/sweep/5206234696cfb37d2d0d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..93fce782373105df421168fc8c8365072864656b
--- /dev/null
+++ b/records/sd1/sweep/5206234696cfb37d2d0d/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5206234696cfb37d2d0d",
+ "values": {
+  "cosine_distance": {
+   "value": 0.550189,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.669813,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.726468,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.322622,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.455,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00052,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 44.816,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0305,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.9022,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5206234696cfb37d2d0d/record.json b/records/sd1/sweep/5206234696cfb37d2d0d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..11e51381ba109589e2aae81971b1c9560ff9013a
--- /dev/null
+++ b/records/sd1/sweep/5206234696cfb37d2d0d/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5206234696cfb37d2d0d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:39:40Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.0.skip_connection",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "in.hi",
+   "kind": "skip",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/507a4ab6691e2276c44ae26b7d9adef10cb925cfacf470b7fb0d8974ab1e3101.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "507a4ab6691e2276c44ae26b7d9adef10cb925cfacf470b7fb0d8974ab1e3101",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "42b45f048d27b771a24d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5206234696cfb37d2d0d/workflow.json b/records/sd1/sweep/5206234696cfb37d2d0d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ae8268b126a9db8184d3d8f2cdb87ad0933c21a2
--- /dev/null
+++ b/records/sd1/sweep/5206234696cfb37d2d0d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5206234696cfb37d2d0d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.0.skip_connection\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5210ed94f094a3cff406/measurements.json b/records/sd1/sweep/5210ed94f094a3cff406/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cf5715c419c117ee925d5f84ec00efbbdaeee7bb
--- /dev/null
+++ b/records/sd1/sweep/5210ed94f094a3cff406/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5210ed94f094a3cff406",
+ "values": {
+  "cosine_distance": {
+   "value": 0.020732,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.033934,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.013044,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.007224,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.466,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00572,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 5.307,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0106,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1202,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5210ed94f094a3cff406/record.json b/records/sd1/sweep/5210ed94f094a3cff406/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ff5d2473f54ea5e6a819775a01e7ff3b5b1c84c8
--- /dev/null
+++ b/records/sd1/sweep/5210ed94f094a3cff406/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5210ed94f094a3cff406",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:43:51Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.1.0.emb_layers.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Linear",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/2276b13595deb40d4f33da54adf7bf8175d7f244f9d994b614b1509341f80924.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "2276b13595deb40d4f33da54adf7bf8175d7f244f9d994b614b1509341f80924",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6cb2a8201229e064f6dd"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5210ed94f094a3cff406/workflow.json b/records/sd1/sweep/5210ed94f094a3cff406/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..546bab39cc598b44e74175e82ad80ff3cd174a0b
--- /dev/null
+++ b/records/sd1/sweep/5210ed94f094a3cff406/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5210ed94f094a3cff406)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.1.0.emb_layers.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52155fdfa4e4539f9812/measurements.json b/records/sd1/sweep/52155fdfa4e4539f9812/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c12dc90e757859836c4d35d8622eb9b363ef1018
--- /dev/null
+++ b/records/sd1/sweep/52155fdfa4e4539f9812/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52155fdfa4e4539f9812",
+ "values": {
+  "std": {
+   "value": 39.859,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0146,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 46.798,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.511773,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.543562,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.248966,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.070803,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0167,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0786,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52155fdfa4e4539f9812/record.json b/records/sd1/sweep/52155fdfa4e4539f9812/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..0e28098b852cbac034a4a44cb6f7ada186e6e534
--- /dev/null
+++ b/records/sd1/sweep/52155fdfa4e4539f9812/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "52155fdfa4e4539f9812",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:38:58Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "zero",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "multiply",
+   0.0,
+   "early"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "4f5728f4228280cb1724"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52155fdfa4e4539f9812/workflow.json b/records/sd1/sweep/52155fdfa4e4539f9812/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..87f5c74fe0bad24cf483ce0f481dbcd3ec25a002
--- /dev/null
+++ b/records/sd1/sweep/52155fdfa4e4539f9812/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52155fdfa4e4539f9812)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/521a0939caea2f17c9ce/measurements.json b/records/sd1/sweep/521a0939caea2f17c9ce/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f295f03bc5101f07da598fcbd2fb1e9000f891b3
--- /dev/null
+++ b/records/sd1/sweep/521a0939caea2f17c9ce/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "521a0939caea2f17c9ce",
+ "values": {
+  "std": {
+   "value": 37.861,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01427,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 4.645,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.016147,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.050696,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.021805,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.005616,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0177,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0138,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/521a0939caea2f17c9ce/record.json b/records/sd1/sweep/521a0939caea2f17c9ce/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..61f50d072cfa9f2736acf0fb244e87a8a1132b95
--- /dev/null
+++ b/records/sd1/sweep/521a0939caea2f17c9ce/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "521a0939caea2f17c9ce",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:07:24Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90.0
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "quarter",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "rotate",
+   90.0,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "0b7ae0bb4ea8131bdd51"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/521a0939caea2f17c9ce/workflow.json b/records/sd1/sweep/521a0939caea2f17c9ce/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d51ca725e868956a7401a35036c7fd3acec7f42
--- /dev/null
+++ b/records/sd1/sweep/521a0939caea2f17c9ce/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 521a0939caea2f17c9ce)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90.0\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5221b65f251873f4fbfc/measurements.json b/records/sd1/sweep/5221b65f251873f4fbfc/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..7bab2eaabb60820492204c18cc987eda74d76f14
--- /dev/null
+++ b/records/sd1/sweep/5221b65f251873f4fbfc/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5221b65f251873f4fbfc",
+ "values": {
+  "cosine_distance": {
+   "value": 0.015975,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.044796,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.005743,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.002925,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.875,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00612,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 6.268,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0255,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.148,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5221b65f251873f4fbfc/record.json b/records/sd1/sweep/5221b65f251873f4fbfc/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..0b8f4f73f9dae28bb6c0ce2c575643e24020989b
--- /dev/null
+++ b/records/sd1/sweep/5221b65f251873f4fbfc/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5221b65f251873f4fbfc",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:51Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.in_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 1.5
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/f1c09e7ce4504db9ec9be09a03860705ca639bc480f3abaf56e3b445bf3a0249.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "f1c09e7ce4504db9ec9be09a03860705ca639bc480f3abaf56e3b445bf3a0249",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "0838f069bdef261032b2"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5221b65f251873f4fbfc/workflow.json b/records/sd1/sweep/5221b65f251873f4fbfc/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..58ca64c2943ef7af7d0469a2fc252e57675ba3c0
--- /dev/null
+++ b/records/sd1/sweep/5221b65f251873f4fbfc/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5221b65f251873f4fbfc)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.in_layers.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5222d9db663159063d37/interpretations.jsonl b/records/sd1/sweep/5222d9db663159063d37/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..23a351597db65a66087a89491c8ab6a61cf9e8a3
--- /dev/null
+++ b/records/sd1/sweep/5222d9db663159063d37/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "5222d9db663159063d37"}, "kind": "caption", "value": "Abstract smeared pale oval shape surrounded by dark expressive brushstrokes on grey", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:19:38Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "00f377f134192cf1"}
+{"target": {"record": "5222d9db663159063d37"}, "kind": "change", "value": "The portrait collapses into a gestural abstract painting: a pinkish oval blob framed by rough black strokes, with no recognisable features.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:19:38Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "295e61b7a67b959e"}
+{"target": {"record": "5222d9db663159063d37"}, "kind": "keywords", "value": ["gestural abstraction", "smeared face", "expressive brushstrokes", "abstract portrait", "dark strokes"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:19:38Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "44bc8089d61db50c"}
+{"target": {"record": "5222d9db663159063d37"}, "kind": "effect_tags", "value": ["abstract", "painterly", "subject_lost", "distorted", "texture"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:19:38Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "790fe163f99a3094"}
diff --git a/records/sd1/sweep/5222d9db663159063d37/measurements.json b/records/sd1/sweep/5222d9db663159063d37/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..eb7b5844dc6273a8c2ce314e56e8e4d48e5dd4f7
--- /dev/null
+++ b/records/sd1/sweep/5222d9db663159063d37/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5222d9db663159063d37",
+ "values": {
+  "cosine_distance": {
+   "value": 0.378631,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.559226,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.681122,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.243901,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 43.446,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0033,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 41.299,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0128,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3591,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5222d9db663159063d37/record.json b/records/sd1/sweep/5222d9db663159063d37/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ae5aa50eca2f6bcb05553de9fac67bb95b423468
--- /dev/null
+++ b/records/sd1/sweep/5222d9db663159063d37/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5222d9db663159063d37",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:23Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.6.0.op",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "in.mid",
+   "kind": "downsample",
+   "module_type": "Conv2d",
+   "container": "Downsample",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/ef26a6852765aebed310edfd6f06eb3dc50d2bfd7e7efef4ab4181344bc9cd17.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "ef26a6852765aebed310edfd6f06eb3dc50d2bfd7e7efef4ab4181344bc9cd17",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.Downsample.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a1423f3f0913085cea61"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5222d9db663159063d37/workflow.json b/records/sd1/sweep/5222d9db663159063d37/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..fe7a9d0088bdcaaec3efc7f157b96c49004322d6
--- /dev/null
+++ b/records/sd1/sweep/5222d9db663159063d37/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5222d9db663159063d37)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.6.0.op\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/522c590c406049f781e1/measurements.json b/records/sd1/sweep/522c590c406049f781e1/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..8ac16fef3df4f2f3365712a6e785c79ac75b36fc
--- /dev/null
+++ b/records/sd1/sweep/522c590c406049f781e1/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "522c590c406049f781e1",
+ "values": {
+  "cosine_distance": {
+   "value": 0.125407,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.211066,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.233221,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.1059,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.783,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0027,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 19.836,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9946,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2064,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/522c590c406049f781e1/record.json b/records/sd1/sweep/522c590c406049f781e1/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1773aad082432f870ee91ab9ba449553d1aac5e9
--- /dev/null
+++ b/records/sd1/sweep/522c590c406049f781e1/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "522c590c406049f781e1",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:21Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.0.in_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/b79ce1a6a1661a46d655ed1d4f8889983b391ed53516f9f04d45d185bdbc7f67.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "b79ce1a6a1661a46d655ed1d4f8889983b391ed53516f9f04d45d185bdbc7f67",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "1041a5faf63a0e59eebc"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/522c590c406049f781e1/workflow.json b/records/sd1/sweep/522c590c406049f781e1/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..6b3600202a9bd1af31584f62fcfa4266e0473f9c
--- /dev/null
+++ b/records/sd1/sweep/522c590c406049f781e1/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 522c590c406049f781e1)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.0.in_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/524262fa099f945a9e66/measurements.json b/records/sd1/sweep/524262fa099f945a9e66/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f205a522115233ed4af021824fa97d3e42fc49dd
--- /dev/null
+++ b/records/sd1/sweep/524262fa099f945a9e66/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "524262fa099f945a9e66",
+ "values": {
+  "cosine_distance": {
+   "value": 0.142711,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.185573,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.053204,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.039107,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.978,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00549,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 18.835,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9931,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.075,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/524262fa099f945a9e66/record.json b/records/sd1/sweep/524262fa099f945a9e66/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..db9b3573914de206113481d0acd581fb0c364a4c
--- /dev/null
+++ b/records/sd1/sweep/524262fa099f945a9e66/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "524262fa099f945a9e66",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:41:41Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.1.transformer_blocks.0.attn2.to_q",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.mid",
+   "kind": "cross_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/012ea4c4f569c808b6d72f8be3bdb65e2b8f82c8873622168b732700fda4fd2a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "012ea4c4f569c808b6d72f8be3bdb65e2b8f82c8873622168b732700fda4fd2a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "35b35908bcb546d3e65a"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/524262fa099f945a9e66/workflow.json b/records/sd1/sweep/524262fa099f945a9e66/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..0172c33d66f878a78ed7b641467a2a40c79c8570
--- /dev/null
+++ b/records/sd1/sweep/524262fa099f945a9e66/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 524262fa099f945a9e66)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.1.transformer_blocks.0.attn2.to_q\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5243d3ef836199f51bf4/measurements.json b/records/sd1/sweep/5243d3ef836199f51bf4/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..b2215bae1e61479948621b51ce808f01596d1bf6
--- /dev/null
+++ b/records/sd1/sweep/5243d3ef836199f51bf4/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5243d3ef836199f51bf4",
+ "values": {
+  "std": {
+   "value": 43.739,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01068,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 27.588,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.3673,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.478373,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.253898,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.066053,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9986,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0472,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5243d3ef836199f51bf4/record.json b/records/sd1/sweep/5243d3ef836199f51bf4/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5bd4546a01c5355c16e4bcd3bda75d1129bbc29c
--- /dev/null
+++ b/records/sd1/sweep/5243d3ef836199f51bf4/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "5243d3ef836199f51bf4",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:40:01Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90.0
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "quarter",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "rotate",
+   90.0,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "614fccee2c0b31f9aa17"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5243d3ef836199f51bf4/workflow.json b/records/sd1/sweep/5243d3ef836199f51bf4/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..fc4d2ac302868a1dbc083e5ac737d99a90e26853
--- /dev/null
+++ b/records/sd1/sweep/5243d3ef836199f51bf4/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5243d3ef836199f51bf4)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/524d2f5861e4f6cf0399/measurements.json b/records/sd1/sweep/524d2f5861e4f6cf0399/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..4323596a79ce76aa6fa51a3c06f14689fc7b034a
--- /dev/null
+++ b/records/sd1/sweep/524d2f5861e4f6cf0399/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "524d2f5861e4f6cf0399",
+ "values": {
+  "cosine_distance": {
+   "value": 0.409599,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.507937,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.309355,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.15833,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 40.133,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00287,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 40.584,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9744,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3487,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/524d2f5861e4f6cf0399/record.json b/records/sd1/sweep/524d2f5861e4f6cf0399/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3cdabf9c6cddacd3fd096797bf9c5b158dec5868
--- /dev/null
+++ b/records/sd1/sweep/524d2f5861e4f6cf0399/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "524d2f5861e4f6cf0399",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:49:17Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.0.0.out_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/8c3e64fc54bcf37db338d9255f17dae41f17b7e5106f9f52441782b16db64cb9.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "8c3e64fc54bcf37db338d9255f17dae41f17b7e5106f9f52441782b16db64cb9",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "bae0209f9cc81eefb350"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/524d2f5861e4f6cf0399/workflow.json b/records/sd1/sweep/524d2f5861e4f6cf0399/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..06ceaa30d1dc497fb1bab30ea393d5e9f487f1bb
--- /dev/null
+++ b/records/sd1/sweep/524d2f5861e4f6cf0399/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 524d2f5861e4f6cf0399)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.0.0.out_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5257c946ec17f033105f/measurements.json b/records/sd1/sweep/5257c946ec17f033105f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..ffce026e954794a34fad404320a2b49669b69ecd
--- /dev/null
+++ b/records/sd1/sweep/5257c946ec17f033105f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5257c946ec17f033105f",
+ "values": {
+  "std": {
+   "value": 72.486,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00797,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 53.215,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.417102,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.432588,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.078629,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.048926,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0116,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3896,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5257c946ec17f033105f/record.json b/records/sd1/sweep/5257c946ec17f033105f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..78c844dda98a8ab403dad2dbc2db296160fda637
--- /dev/null
+++ b/records/sd1/sweep/5257c946ec17f033105f/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "5257c946ec17f033105f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:34:00Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "grow",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "scale",
+   1.4,
+   "early"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 1.4,
+    "applied": 1.4,
+    "reason": "outside safe range [0.95, 1.05] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "16d5d95df4f8602a1867"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5257c946ec17f033105f/workflow.json b/records/sd1/sweep/5257c946ec17f033105f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..e8b53897b129d501824053ee0d404e9af3ae2e24
--- /dev/null
+++ b/records/sd1/sweep/5257c946ec17f033105f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5257c946ec17f033105f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/525fd84622520b0f8545/measurements.json b/records/sd1/sweep/525fd84622520b0f8545/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..14bd8cb0b4823c746bf8870c53fc0679ec80a2b7
--- /dev/null
+++ b/records/sd1/sweep/525fd84622520b0f8545/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "525fd84622520b0f8545",
+ "values": {
+  "cosine_distance": {
+   "value": 0.250176,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.326754,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.153061,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.06359,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 50.272,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00522,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 30.406,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.973,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0668,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/525fd84622520b0f8545/record.json b/records/sd1/sweep/525fd84622520b0f8545/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1cd58b45e3bf7aef8fcc4b47fd607cfb3b7cb794
--- /dev/null
+++ b/records/sd1/sweep/525fd84622520b0f8545/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "525fd84622520b0f8545",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:33Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.7.1.proj_out",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "in.lo",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/299ada7011e37160a931760ae3d66fff729394293171bf424700fdbc5ebf7ec9.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "299ada7011e37160a931760ae3d66fff729394293171bf424700fdbc5ebf7ec9",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "616e3e3568b3d7d2d7f4"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/525fd84622520b0f8545/workflow.json b/records/sd1/sweep/525fd84622520b0f8545/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..97e0868b0e206f8fa6aef71ca02c3d18967ce73b
--- /dev/null
+++ b/records/sd1/sweep/525fd84622520b0f8545/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 525fd84622520b0f8545)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.7.1.proj_out\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5263bd86371d30758366/measurements.json b/records/sd1/sweep/5263bd86371d30758366/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..4df246b43808f19eb4c6ef8ea9d3a4487003ca67
--- /dev/null
+++ b/records/sd1/sweep/5263bd86371d30758366/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5263bd86371d30758366",
+ "values": {
+  "cosine_distance": {
+   "value": 0.024224,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.036879,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.013625,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.007563,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.667,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00538,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 6.051,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.99,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1505,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5263bd86371d30758366/record.json b/records/sd1/sweep/5263bd86371d30758366/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..86de83d1cfcd8712d0e6d82042a9bf4ca402398b
--- /dev/null
+++ b/records/sd1/sweep/5263bd86371d30758366/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5263bd86371d30758366",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:43:55Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.1.0.skip_connection",
+   "op": "multiply",
+   "args": {
+    "scalar": 1.5
+   },
+   "group": "out.lo",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/402cef2883ccf14f2bc9f8e3da58cec759fc5dd087a87ebbe471be7e89adef3e.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "402cef2883ccf14f2bc9f8e3da58cec759fc5dd087a87ebbe471be7e89adef3e",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a94d5388d58e486ea8d6"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5263bd86371d30758366/workflow.json b/records/sd1/sweep/5263bd86371d30758366/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..1c214e3914f06374f46f62a8a4d442703a46b9ff
--- /dev/null
+++ b/records/sd1/sweep/5263bd86371d30758366/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5263bd86371d30758366)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.1.0.skip_connection\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5276999ce68066474239/measurements.json b/records/sd1/sweep/5276999ce68066474239/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e5f0c04528c2c7845de2bf370c7d099098853891
--- /dev/null
+++ b/records/sd1/sweep/5276999ce68066474239/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5276999ce68066474239",
+ "values": {
+  "std": {
+   "value": 28.637,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01619,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 14.754,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.07854,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.256733,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.350415,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.064537,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0133,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2726,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5276999ce68066474239/record.json b/records/sd1/sweep/5276999ce68066474239/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9237ba96b7ca11fbb5a2fda3aca22b7b5fdd7c8d
--- /dev/null
+++ b/records/sd1/sweep/5276999ce68066474239/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "5276999ce68066474239",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:16:56Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "grow",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "scale",
+   1.4,
+   "late"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 1.4,
+    "applied": 1.4,
+    "reason": "outside safe range [0.7, 1.3] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "c61d9783921ce68bf1d3"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5276999ce68066474239/workflow.json b/records/sd1/sweep/5276999ce68066474239/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..3e1b5b8c5576f0006179c53466fd16b80a3ff6fb
--- /dev/null
+++ b/records/sd1/sweep/5276999ce68066474239/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5276999ce68066474239)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52811374c7b1219efb56/measurements.json b/records/sd1/sweep/52811374c7b1219efb56/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..5a2319f0659a082bcf8b42689ac63d0fc9a37e2d
--- /dev/null
+++ b/records/sd1/sweep/52811374c7b1219efb56/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52811374c7b1219efb56",
+ "values": {
+  "std": {
+   "value": 41.809,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01879,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 50.287,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.563548,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.559375,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.165555,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.095273,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0226,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1274,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52811374c7b1219efb56/record.json b/records/sd1/sweep/52811374c7b1219efb56/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..f8182f059f0e8eee2ec06c3dbeebdd6ecd359bda
--- /dev/null
+++ b/records/sd1/sweep/52811374c7b1219efb56/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "52811374c7b1219efb56",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:39:15Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "zero",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "multiply",
+   0.0,
+   "early"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "aed20cf7156b17adb1ae"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52811374c7b1219efb56/workflow.json b/records/sd1/sweep/52811374c7b1219efb56/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7e765d1b34ad4d235b30b3963371345d32d5e536
--- /dev/null
+++ b/records/sd1/sweep/52811374c7b1219efb56/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52811374c7b1219efb56)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5297d16c9ba5846e5e40/measurements.json b/records/sd1/sweep/5297d16c9ba5846e5e40/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..46b9a7521f6e8a08e656cc7af3f4d6512267196c
--- /dev/null
+++ b/records/sd1/sweep/5297d16c9ba5846e5e40/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5297d16c9ba5846e5e40",
+ "values": {
+  "std": {
+   "value": 54.778,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00352,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 21.441,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.158866,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.446688,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.322841,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.181839,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9478,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.169,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5297d16c9ba5846e5e40/record.json b/records/sd1/sweep/5297d16c9ba5846e5e40/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..45b3b9216d14a808522d6ab5e4b83d5c9ef5a368
--- /dev/null
+++ b/records/sd1/sweep/5297d16c9ba5846e5e40/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "5297d16c9ba5846e5e40",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:34:05Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "grow",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "scale",
+   1.4,
+   "late"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 1.4,
+    "applied": 1.4,
+    "reason": "outside safe range [0.95, 1.05] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "8d6a9e17ebf1937be495"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5297d16c9ba5846e5e40/workflow.json b/records/sd1/sweep/5297d16c9ba5846e5e40/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7116952101f96d6f2d01a7eee8d54d9ac1c32bb3
--- /dev/null
+++ b/records/sd1/sweep/5297d16c9ba5846e5e40/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5297d16c9ba5846e5e40)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52aa9a44de782862457f/measurements.json b/records/sd1/sweep/52aa9a44de782862457f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c53244cf9a49d85cd5223610bdc4d9bca4e477ba
--- /dev/null
+++ b/records/sd1/sweep/52aa9a44de782862457f/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "52aa9a44de782862457f",
+ "values": {
+  "cosine_distance": {
+   "value": 0.38144,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.450923,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.221601,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.162164,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.045,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00383,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 44.715,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.024,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1311,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52aa9a44de782862457f/record.json b/records/sd1/sweep/52aa9a44de782862457f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..4275b4692b9b0144d8ce77b66321c1803b5e3dc0
--- /dev/null
+++ b/records/sd1/sweep/52aa9a44de782862457f/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "52aa9a44de782862457f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:45Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.7.1.proj_in",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.5
+   },
+   "group": "in.lo",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/e92bbf4dd9ce32a64551167a9b46a3c3c0efb4b0e0766409ca184da0e414a580.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "e92bbf4dd9ce32a64551167a9b46a3c3c0efb4b0e0766409ca184da0e414a580",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "d90f0c5894dbfd823ece"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52aa9a44de782862457f/workflow.json b/records/sd1/sweep/52aa9a44de782862457f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a3dcfa208e1fe1542aee49ff48fb0fc68d224100
--- /dev/null
+++ b/records/sd1/sweep/52aa9a44de782862457f/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52aa9a44de782862457f)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.7.1.proj_in\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52ae5aae42334f1c9109/measurements.json b/records/sd1/sweep/52ae5aae42334f1c9109/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..b6a78f68b2d140bc991c08dc20fb1b02b00fa3d1
--- /dev/null
+++ b/records/sd1/sweep/52ae5aae42334f1c9109/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52ae5aae42334f1c9109",
+ "values": {
+  "std": {
+   "value": 37.088,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02596,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 42.437,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.47362,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.446102,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.356025,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.057241,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.03,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1212,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52ae5aae42334f1c9109/record.json b/records/sd1/sweep/52ae5aae42334f1c9109/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..bd0efd42a569ce38cf620fb71c6de11c5d200bae
--- /dev/null
+++ b/records/sd1/sweep/52ae5aae42334f1c9109/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "52ae5aae42334f1c9109",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:01:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "zero",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "13216623a742ccb32c5e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "multiply",
+   0.0,
+   "early"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "cfa0754348160243265a"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52ae5aae42334f1c9109/workflow.json b/records/sd1/sweep/52ae5aae42334f1c9109/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c723b1fafe98191fa684d59273968425bcae321a
--- /dev/null
+++ b/records/sd1/sweep/52ae5aae42334f1c9109/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52ae5aae42334f1c9109)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52b373b6b6df856956eb/measurements.json b/records/sd1/sweep/52b373b6b6df856956eb/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..66e5ce95581b5e11b70e3286a74aa99da9153817
--- /dev/null
+++ b/records/sd1/sweep/52b373b6b6df856956eb/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "52b373b6b6df856956eb",
+ "values": {
+  "cosine_distance": {
+   "value": 0.228497,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.223091,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.190484,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.19095,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 64.97,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00576,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 24.306,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9659,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.228,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52b373b6b6df856956eb/record.json b/records/sd1/sweep/52b373b6b6df856956eb/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b1cc043f459e94be7e646baa88cb3d745f462ab5
--- /dev/null
+++ b/records/sd1/sweep/52b373b6b6df856956eb/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "52b373b6b6df856956eb",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:49Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.10.0.in_layers.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 2
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "strong_grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/4db5b3e405991f3fa2b9af810b2e0f2203631836a5d891533cc50fab44e606c1.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "4db5b3e405991f3fa2b9af810b2e0f2203631836a5d891533cc50fab44e606c1",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "aaea08be1584031a0ad9"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52b373b6b6df856956eb/workflow.json b/records/sd1/sweep/52b373b6b6df856956eb/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..67cc90c3d3609d76ff0215614feb7366d3142a5f
--- /dev/null
+++ b/records/sd1/sweep/52b373b6b6df856956eb/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52b373b6b6df856956eb)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.10.0.in_layers.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52b477d84afcb70d4b46/measurements.json b/records/sd1/sweep/52b477d84afcb70d4b46/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..bfe7368bff5777588a43ded14dae597b6cedd472
--- /dev/null
+++ b/records/sd1/sweep/52b477d84afcb70d4b46/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52b477d84afcb70d4b46",
+ "values": {
+  "cosine_distance": {
+   "value": 0.268191,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.429839,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.249247,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.14764,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 41.751,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00477,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 29.022,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9537,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2251,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52b477d84afcb70d4b46/record.json b/records/sd1/sweep/52b477d84afcb70d4b46/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..149d51dd6ddf8accfd50b5985bd767c567e39bf8
--- /dev/null
+++ b/records/sd1/sweep/52b477d84afcb70d4b46/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "52b477d84afcb70d4b46",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:24Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.5.1.proj_out",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.lo",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/17409124eb95b2f8671b4e93aae7ba397b36d22602109832784534471d792b8e.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "17409124eb95b2f8671b4e93aae7ba397b36d22602109832784534471d792b8e",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "9ff2c457d082ba4ad6c4"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52b477d84afcb70d4b46/workflow.json b/records/sd1/sweep/52b477d84afcb70d4b46/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..da40020575f990f936b3261b6a0ec32e713140dd
--- /dev/null
+++ b/records/sd1/sweep/52b477d84afcb70d4b46/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52b477d84afcb70d4b46)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.5.1.proj_out\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52ca38acabc043ec7529/measurements.json b/records/sd1/sweep/52ca38acabc043ec7529/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cb8a935b6e5564c601fb273236da9f0877196323
--- /dev/null
+++ b/records/sd1/sweep/52ca38acabc043ec7529/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "52ca38acabc043ec7529",
+ "values": {
+  "cosine_distance": {
+   "value": 0.162803,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.325412,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.220385,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.100812,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.023,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00396,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 21.48,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0061,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0727,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52ca38acabc043ec7529/record.json b/records/sd1/sweep/52ca38acabc043ec7529/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d0aa904023b8bef57bab94bf28526ba47204240a
--- /dev/null
+++ b/records/sd1/sweep/52ca38acabc043ec7529/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "52ca38acabc043ec7529",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:50:39Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.0.out_layers.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/997974d48b7cd2d9168824b848346744121abfed71e4302f70077361d5e6c7a9.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "997974d48b7cd2d9168824b848346744121abfed71e4302f70077361d5e6c7a9",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "8e2a06f1ce75a17289b9"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52ca38acabc043ec7529/workflow.json b/records/sd1/sweep/52ca38acabc043ec7529/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f6d0ded15c9825f23fa0d8a1a9145f4337dbe92e
--- /dev/null
+++ b/records/sd1/sweep/52ca38acabc043ec7529/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52ca38acabc043ec7529)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.0.out_layers.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/measurements.json b/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6c0e37a18d0a109b9012e350a4569ed7a537ed6f
--- /dev/null
+++ b/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52cd27ddcc0dd7ecd0b8",
+ "values": {
+  "std": {
+   "value": 49.912,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.009,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 10.433,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.06753,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.209406,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.18954,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.05213,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0828,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.004,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/record.json b/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..43ff050220172cb2fb6acfbd4c61317864d0506c
--- /dev/null
+++ b/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "52cd27ddcc0dd7ecd0b8",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:37:33Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.7
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "shrink",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "scale",
+   0.7,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "8b72d5a5f2a53f1c63a4"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/workflow.json b/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f6ddd1e938423ceef45e1eff7da2a356826760f3
--- /dev/null
+++ b/records/sd1/sweep/52cd27ddcc0dd7ecd0b8/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52cd27ddcc0dd7ecd0b8)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.7\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d00c82baf9f257bf15/measurements.json b/records/sd1/sweep/52d00c82baf9f257bf15/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..33e84d59d3f011d67a99af5e4949c1bb1bec3ba3
--- /dev/null
+++ b/records/sd1/sweep/52d00c82baf9f257bf15/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52d00c82baf9f257bf15",
+ "values": {
+  "std": {
+   "value": 22.548,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01825,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 14.423,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.094776,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.379555,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.862757,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.307579,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.8478,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4339,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d00c82baf9f257bf15/record.json b/records/sd1/sweep/52d00c82baf9f257bf15/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ac675da768aa896762eaea7ad80028a915489e73
--- /dev/null
+++ b/records/sd1/sweep/52d00c82baf9f257bf15/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "52d00c82baf9f257bf15",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:52:50Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "zero",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "multiply",
+   0.0,
+   "late"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "5a58f1cc1fd520728faf"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d00c82baf9f257bf15/workflow.json b/records/sd1/sweep/52d00c82baf9f257bf15/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..3ccdc1717db67d8ae5846a6ac78a4a21b0b47ff2
--- /dev/null
+++ b/records/sd1/sweep/52d00c82baf9f257bf15/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52d00c82baf9f257bf15)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d2b2b568cad1d7d10c/measurements.json b/records/sd1/sweep/52d2b2b568cad1d7d10c/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..88ef353445421004b80f60086d9b836eb0d93292
--- /dev/null
+++ b/records/sd1/sweep/52d2b2b568cad1d7d10c/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52d2b2b568cad1d7d10c",
+ "values": {
+  "cosine_distance": {
+   "value": 0.192409,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.192408,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.129721,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.069431,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 67.072,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00646,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 20.685,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9641,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2361,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d2b2b568cad1d7d10c/record.json b/records/sd1/sweep/52d2b2b568cad1d7d10c/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ed9b7f4584ba8c1909923cacd7a0c5735cd6d1d2
--- /dev/null
+++ b/records/sd1/sweep/52d2b2b568cad1d7d10c/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "52d2b2b568cad1d7d10c",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:20Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.1.norm",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "in.mid",
+   "kind": "norm",
+   "module_type": "GroupNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/03752f3f254a89dca3e47651f20ef0b05d94e559ed50f3f9e6dd67b83de7e7d7.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "03752f3f254a89dca3e47651f20ef0b05d94e559ed50f3f9e6dd67b83de7e7d7",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "e9984efc4c0e3061e0e0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d2b2b568cad1d7d10c/workflow.json b/records/sd1/sweep/52d2b2b568cad1d7d10c/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..239a8cb18c596fe2dc6c4612079fd6b44b088349
--- /dev/null
+++ b/records/sd1/sweep/52d2b2b568cad1d7d10c/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52d2b2b568cad1d7d10c)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.1.norm\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d5d0075e49eab1c2ea/interpretations.jsonl b/records/sd1/sweep/52d5d0075e49eab1c2ea/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..f35f651b50bff4cd6a6222e819abb32bc52ea97c
--- /dev/null
+++ b/records/sd1/sweep/52d5d0075e49eab1c2ea/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "52d5d0075e49eab1c2ea"}, "kind": "caption", "value": "Short-haired woman in black shirt, grey backdrop with cream vintage border", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:14:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "f627a69d6b823f74"}
+{"target": {"record": "52d5d0075e49eab1c2ea"}, "kind": "change", "value": "Almost no visible change: the same portrait with slightly altered features and a slightly more visible print border.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:14:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a0e7368bb737fa75"}
+{"target": {"record": "52d5d0075e49eab1c2ea"}, "kind": "keywords", "value": ["near identical", "studio portrait", "print border", "minor variation"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:14:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3684fd4c773518c3"}
+{"target": {"record": "52d5d0075e49eab1c2ea"}, "kind": "effect_tags", "value": ["subject_kept", "subtle"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:14:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a4c2f6e379d1142b"}
diff --git a/records/sd1/sweep/52d5d0075e49eab1c2ea/measurements.json b/records/sd1/sweep/52d5d0075e49eab1c2ea/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..d679b29c6da867b0ac7492933e7e7eb9b442d17c
--- /dev/null
+++ b/records/sd1/sweep/52d5d0075e49eab1c2ea/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "52d5d0075e49eab1c2ea",
+ "values": {
+  "cosine_distance": {
+   "value": 0.127473,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.150323,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.081382,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.067931,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.148,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00468,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 17.058,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9363,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.617,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d5d0075e49eab1c2ea/record.json b/records/sd1/sweep/52d5d0075e49eab1c2ea/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..20116ef2b55bf43253185bdcbba3b30e9ac6362d
--- /dev/null
+++ b/records/sd1/sweep/52d5d0075e49eab1c2ea/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "52d5d0075e49eab1c2ea",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:41Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "middle_block.2.in_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/bc7795c5f3a4d71e3ce9a927d9905c1ac65dfc105da3101378954562b63e9193.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "bc7795c5f3a4d71e3ce9a927d9905c1ac65dfc105da3101378954562b63e9193",
+  "type_path": "UNetModel.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "e213f201f6e12c5f4ce5"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d5d0075e49eab1c2ea/workflow.json b/records/sd1/sweep/52d5d0075e49eab1c2ea/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9864dc9eb7f8af1f0fce54b93519ed74cd2fa625
--- /dev/null
+++ b/records/sd1/sweep/52d5d0075e49eab1c2ea/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52d5d0075e49eab1c2ea)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.2.in_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d991eb59dbf9f0b6b8/measurements.json b/records/sd1/sweep/52d991eb59dbf9f0b6b8/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..b50e68d1dddef130dcc7c4ce293a8f0e0e35b86b
--- /dev/null
+++ b/records/sd1/sweep/52d991eb59dbf9f0b6b8/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52d991eb59dbf9f0b6b8",
+ "values": {
+  "cosine_distance": {
+   "value": 0.145224,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.251491,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.109647,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.035386,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 57.913,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00706,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 21.304,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0164,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.071,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d991eb59dbf9f0b6b8/record.json b/records/sd1/sweep/52d991eb59dbf9f0b6b8/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..947eb5a63db30956cbaef741ffebfcd2d5e779ab
--- /dev/null
+++ b/records/sd1/sweep/52d991eb59dbf9f0b6b8/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "52d991eb59dbf9f0b6b8",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:50Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.1.transformer_blocks.0.norm3",
+   "op": "multiply",
+   "args": {
+    "scalar": 1.5
+   },
+   "group": "out.mid",
+   "kind": "norm",
+   "module_type": "LayerNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/4b1f5f60e27c6fafd577b7c105c98e36fcc65d3fa253b7dc6a6205516e2e037e.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "4b1f5f60e27c6fafd577b7c105c98e36fcc65d3fa253b7dc6a6205516e2e037e",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.LayerNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "5df013e7c4cba886a32b"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52d991eb59dbf9f0b6b8/workflow.json b/records/sd1/sweep/52d991eb59dbf9f0b6b8/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..bc94c182bc9845a4a04a6d4611fd35c7d34c9185
--- /dev/null
+++ b/records/sd1/sweep/52d991eb59dbf9f0b6b8/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52d991eb59dbf9f0b6b8)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.1.transformer_blocks.0.norm3\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52e37b2c4c32eed03b55/measurements.json b/records/sd1/sweep/52e37b2c4c32eed03b55/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..733d46e5edb46c5c3bf7204f816e7166faab1c16
--- /dev/null
+++ b/records/sd1/sweep/52e37b2c4c32eed03b55/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "52e37b2c4c32eed03b55",
+ "values": {
+  "cosine_distance": {
+   "value": 0.318977,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.33282,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.404283,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.299532,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.767,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00524,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 35.68,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8718,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5924,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52e37b2c4c32eed03b55/record.json b/records/sd1/sweep/52e37b2c4c32eed03b55/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..807436e20bb83990b54fbb4883d914272cb6b5a2
--- /dev/null
+++ b/records/sd1/sweep/52e37b2c4c32eed03b55/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "52e37b2c4c32eed03b55",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:48:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.1.transformer_blocks.0.attn2.to_q",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "in.mid",
+   "kind": "cross_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/0d26c313656be274d7ccb4e3bf2446087f437b055448e078e59d7b6395627e6a.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "0d26c313656be274d7ccb4e3bf2446087f437b055448e078e59d7b6395627e6a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "ac66d27a639dbddc188d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52e37b2c4c32eed03b55/workflow.json b/records/sd1/sweep/52e37b2c4c32eed03b55/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..11672777ee7aa6ca4d37454747c875be7a19d7d2
--- /dev/null
+++ b/records/sd1/sweep/52e37b2c4c32eed03b55/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52e37b2c4c32eed03b55)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.1.transformer_blocks.0.attn2.to_q\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52f435d04825e9006f6c/interpretations.jsonl b/records/sd1/sweep/52f435d04825e9006f6c/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..8bbb13afcb62f56e47bc62a28e019c3e36f0d8f6
--- /dev/null
+++ b/records/sd1/sweep/52f435d04825e9006f6c/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "52f435d04825e9006f6c"}, "kind": "caption", "value": "Close-up of a smiling woman with hair tied back in teal top", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:14:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "5287616489794a99"}
+{"target": {"record": "52f435d04825e9006f6c"}, "kind": "change", "value": "The crop tightens on a warmer, softly smiling woman with tied-back hair and a teal top, so the image feels more colourful and approachable.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:14:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "0d1a2efa53aca99e"}
+{"target": {"record": "52f435d04825e9006f6c"}, "kind": "keywords", "value": ["close-up", "teal top", "warm portrait", "slight smile"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:14:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a4d9d3f8188d1a23"}
+{"target": {"record": "52f435d04825e9006f6c"}, "kind": "effect_tags", "value": ["zoom", "colour_shift", "new_composition"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:14:08Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "95d7eafb22b98337"}
diff --git a/records/sd1/sweep/52f435d04825e9006f6c/measurements.json b/records/sd1/sweep/52f435d04825e9006f6c/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..36e2b88c62e2fd6cc6f6840d0434475a43b4f5cb
--- /dev/null
+++ b/records/sd1/sweep/52f435d04825e9006f6c/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "52f435d04825e9006f6c",
+ "values": {
+  "cosine_distance": {
+   "value": 0.309695,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.382376,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.177708,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.093401,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.195,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00381,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 32.552,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9998,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2654,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52f435d04825e9006f6c/record.json b/records/sd1/sweep/52f435d04825e9006f6c/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..fe3b993cd35fb4e2084552618aececac748e04e7
--- /dev/null
+++ b/records/sd1/sweep/52f435d04825e9006f6c/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "52f435d04825e9006f6c",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:53Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "middle_block.0.in_layers.2",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.5
+   },
+   "group": "mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/66cdc4951a48d0a6376aceeceac0dac98c29854ec69f5cef06fec6b0596a1fba.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "66cdc4951a48d0a6376aceeceac0dac98c29854ec69f5cef06fec6b0596a1fba",
+  "type_path": "UNetModel.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "d7461f33f5e3b1beff0b"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52f435d04825e9006f6c/workflow.json b/records/sd1/sweep/52f435d04825e9006f6c/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..51523cea7e9eda806cfe108922328e22a3667ffa
--- /dev/null
+++ b/records/sd1/sweep/52f435d04825e9006f6c/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52f435d04825e9006f6c)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.0.in_layers.2\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52f649e2e5a477757400/measurements.json b/records/sd1/sweep/52f649e2e5a477757400/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cc685fe225bbd7f1c396d6e87b902a1ced77be6d
--- /dev/null
+++ b/records/sd1/sweep/52f649e2e5a477757400/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "52f649e2e5a477757400",
+ "values": {
+  "cosine_distance": {
+   "value": 0.111463,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.187258,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.311879,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.049853,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 76.289,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01179,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.492,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9722,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3589,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52f649e2e5a477757400/record.json b/records/sd1/sweep/52f649e2e5a477757400/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..eb3a09f9f4d915d3d4be1fdca38af9033586fa5e
--- /dev/null
+++ b/records/sd1/sweep/52f649e2e5a477757400/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "52f649e2e5a477757400",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:47:03Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.0.out_layers.3",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/e7630ab325c9ee933bd6d4a89a1154bf73861e5c1d4d36a2346e1ad0e8cf2b7e.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "e7630ab325c9ee933bd6d4a89a1154bf73861e5c1d4d36a2346e1ad0e8cf2b7e",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "8f740cf0eab43dc43732"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/52f649e2e5a477757400/workflow.json b/records/sd1/sweep/52f649e2e5a477757400/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c246e17dded339db34fffda2096b530cb884974a
--- /dev/null
+++ b/records/sd1/sweep/52f649e2e5a477757400/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 52f649e2e5a477757400)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.0.out_layers.3\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53067b0da4bd37712278/measurements.json b/records/sd1/sweep/53067b0da4bd37712278/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6e7ffa58bbc92c917200a8b87478b0d551af8e6d
--- /dev/null
+++ b/records/sd1/sweep/53067b0da4bd37712278/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "53067b0da4bd37712278",
+ "values": {
+  "cosine_distance": {
+   "value": 0.552468,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.588487,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.622521,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.226783,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 35.187,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00235,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 49.2,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.063,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1655,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53067b0da4bd37712278/record.json b/records/sd1/sweep/53067b0da4bd37712278/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d93ebe293c1274489aa03a07ee28cef2140900ca
--- /dev/null
+++ b/records/sd1/sweep/53067b0da4bd37712278/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "53067b0da4bd37712278",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:49:37Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.0.in_layers.2",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/f0170628682775d49b0f5222a379cfc85463f44ab460cb6905ed9d0aa0facf42.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "f0170628682775d49b0f5222a379cfc85463f44ab460cb6905ed9d0aa0facf42",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "c10b8ea45f9a3ff46e81"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53067b0da4bd37712278/workflow.json b/records/sd1/sweep/53067b0da4bd37712278/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c4f89b1b97df1821fe921a5fc57859c11163c89e
--- /dev/null
+++ b/records/sd1/sweep/53067b0da4bd37712278/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53067b0da4bd37712278)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.0.in_layers.2\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53165e7d0e8f18cd2dc3/measurements.json b/records/sd1/sweep/53165e7d0e8f18cd2dc3/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..7928067007a6ddfc603184031731adc1e90d4f5a
--- /dev/null
+++ b/records/sd1/sweep/53165e7d0e8f18cd2dc3/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "53165e7d0e8f18cd2dc3",
+ "values": {
+  "std": {
+   "value": 31.821,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02638,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 30.892,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.241803,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.395941,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.120263,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.034967,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9727,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3463,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53165e7d0e8f18cd2dc3/record.json b/records/sd1/sweep/53165e7d0e8f18cd2dc3/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..8fb6bf6e35580ddc433c77ec1bc5bee2fe5c964b
--- /dev/null
+++ b/records/sd1/sweep/53165e7d0e8f18cd2dc3/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "53165e7d0e8f18cd2dc3",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:01:24Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90.0
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "quarter",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "rotate",
+   90.0,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "87fc69a3ca92f365ff3e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53165e7d0e8f18cd2dc3/workflow.json b/records/sd1/sweep/53165e7d0e8f18cd2dc3/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..168d7fb9ba74f4ae3430492cf6d62047dbe2f8a1
--- /dev/null
+++ b/records/sd1/sweep/53165e7d0e8f18cd2dc3/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53165e7d0e8f18cd2dc3)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5317b9f36f98eeead586/measurements.json b/records/sd1/sweep/5317b9f36f98eeead586/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..ef075cab05312ee341b2f2633869b1aefd931a47
--- /dev/null
+++ b/records/sd1/sweep/5317b9f36f98eeead586/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5317b9f36f98eeead586",
+ "values": {
+  "std": {
+   "value": 51.465,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01042,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 3.651,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.010308,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.049751,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.016679,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.005016,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0152,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0116,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5317b9f36f98eeead586/record.json b/records/sd1/sweep/5317b9f36f98eeead586/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5ca816eb50d8f4b5a119ffdbd6dca52f4539347
--- /dev/null
+++ b/records/sd1/sweep/5317b9f36f98eeead586/record.json
@@ -0,0 +1,79 @@
+{
+ "id": "5317b9f36f98eeead586",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:37:56Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.2,
+    "seed": 1234
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "medium",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "add_noise",
+   0.2,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "2042782cd84a95c37ec9"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5317b9f36f98eeead586/workflow.json b/records/sd1/sweep/5317b9f36f98eeead586/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..1456bf48e564a095126ab4945389fa44d046636b
--- /dev/null
+++ b/records/sd1/sweep/5317b9f36f98eeead586/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5317b9f36f98eeead586)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.2,\n    \"seed\": 1234\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53196defcb513b73da4b/interpretations.jsonl b/records/sd1/sweep/53196defcb513b73da4b/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..428ee0b4a5e8e68b9afbd5b0ed211ec9b133cfb5
--- /dev/null
+++ b/records/sd1/sweep/53196defcb513b73da4b/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "53196defcb513b73da4b"}, "kind": "caption", "value": "Young person with short tucked hair and freckles, slight head turn, worn vintage frame", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:21:04Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "dba605933a2edb95"}
+{"target": {"record": "53196defcb513b73da4b"}, "kind": "change", "value": "The composition is about the same, but the face becomes more androgynous and freckled, the hair is tucked behind the ear, and the border edges look a little more aged.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:21:04Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3964ddd9eb228809"}
+{"target": {"record": "53196defcb513b73da4b"}, "kind": "keywords", "value": ["freckles", "androgynous", "portrait variation", "tucked hair", "vintage frame"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:21:04Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a03f1f1ba687e6e4"}
+{"target": {"record": "53196defcb513b73da4b"}, "kind": "effect_tags", "value": ["subject_kept", "subtle"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:21:04Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "47a2775c1702133a"}
diff --git a/records/sd1/sweep/53196defcb513b73da4b/measurements.json b/records/sd1/sweep/53196defcb513b73da4b/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f42cb07f6e58b108dd1e2f3f2f1510d99b068c26
--- /dev/null
+++ b/records/sd1/sweep/53196defcb513b73da4b/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "53196defcb513b73da4b",
+ "values": {
+  "cosine_distance": {
+   "value": 0.139248,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.151354,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.11142,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.129892,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.653,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00539,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 17.993,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9326,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4427,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53196defcb513b73da4b/record.json b/records/sd1/sweep/53196defcb513b73da4b/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3f750325436b7b41c839325881770fcb83d2f3f6
--- /dev/null
+++ b/records/sd1/sweep/53196defcb513b73da4b/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "53196defcb513b73da4b",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:11Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.1.proj_in",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "out.mid",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/a9ca1937899c09d050ddc401f33ad0b7de3d5ccd7882c41bfe395eecfc33b53b.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "a9ca1937899c09d050ddc401f33ad0b7de3d5ccd7882c41bfe395eecfc33b53b",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "47b540e3ae6b190351b0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53196defcb513b73da4b/workflow.json b/records/sd1/sweep/53196defcb513b73da4b/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..cb28c2177ddbe51e6feccf3ed0872151a480b7f6
--- /dev/null
+++ b/records/sd1/sweep/53196defcb513b73da4b/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53196defcb513b73da4b)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.1.proj_in\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5319fa7e6d331b544fc6/measurements.json b/records/sd1/sweep/5319fa7e6d331b544fc6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..59e2155fd83bc3da7513282385cd11eedc7f4f2b
--- /dev/null
+++ b/records/sd1/sweep/5319fa7e6d331b544fc6/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5319fa7e6d331b544fc6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.420662,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.593277,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.350463,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.267001,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 35.005,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0039,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 40.117,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0608,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7123,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5319fa7e6d331b544fc6/record.json b/records/sd1/sweep/5319fa7e6d331b544fc6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9d2a5cbb9c264d5790ead937bc2cea4e254006a8
--- /dev/null
+++ b/records/sd1/sweep/5319fa7e6d331b544fc6/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5319fa7e6d331b544fc6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:49:48Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.4.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/dd945d8277d75f919c5db7ffa54ea4557277e4a420db592ec5ea983e3c2a6353.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "dd945d8277d75f919c5db7ffa54ea4557277e4a420db592ec5ea983e3c2a6353",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a8a92e593bdb06b39728"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5319fa7e6d331b544fc6/workflow.json b/records/sd1/sweep/5319fa7e6d331b544fc6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..03d3033050a225c2fdaeb5f1b0bb0129b13caee2
--- /dev/null
+++ b/records/sd1/sweep/5319fa7e6d331b544fc6/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5319fa7e6d331b544fc6)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.4.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53246b5aafd59634d424/interpretations.jsonl b/records/sd1/sweep/53246b5aafd59634d424/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e4201cfd5aad11076553498ca9d9ce40f899f2cc
--- /dev/null
+++ b/records/sd1/sweep/53246b5aafd59634d424/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "53246b5aafd59634d424"}, "kind": "caption", "value": "Young person with tousled dark hair inside a mottled green painted oval vignette", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:07:29Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "b97ea9778e4431ae"}
+{"target": {"record": "53246b5aafd59634d424"}, "kind": "change", "value": "The portrait becomes a vintage painted look with a tousled-haired figure framed by a weathered green oval vignette and darker distressed edges.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:07:29Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "5655eb18477016a3"}
+{"target": {"record": "53246b5aafd59634d424"}, "kind": "keywords", "value": ["oval vignette", "green painted background", "vintage portrait", "tousled hair", "distressed edges"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:07:29Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a3614b14648b8d03"}
+{"target": {"record": "53246b5aafd59634d424"}, "kind": "effect_tags", "value": ["framing", "painterly", "texture", "colour_shift", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:07:29Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ef8acec65716486a"}
diff --git a/records/sd1/sweep/53246b5aafd59634d424/measurements.json b/records/sd1/sweep/53246b5aafd59634d424/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..8afaba7377964202b3755ca73570a8c803006286
--- /dev/null
+++ b/records/sd1/sweep/53246b5aafd59634d424/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "53246b5aafd59634d424",
+ "values": {
+  "cosine_distance": {
+   "value": 0.4791,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.47319,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.237443,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.133138,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 52.612,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00502,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 46.575,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9757,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0401,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53246b5aafd59634d424/record.json b/records/sd1/sweep/53246b5aafd59634d424/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a2f1ffef12d77ef96312589374284c5c7e538e3c
--- /dev/null
+++ b/records/sd1/sweep/53246b5aafd59634d424/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "53246b5aafd59634d424",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:05Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.0.in_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "in.hi",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/22c8e713759a848d2bac0f0a2d40cf82a8eae9ef51155d0f220890d8f790dc56.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "22c8e713759a848d2bac0f0a2d40cf82a8eae9ef51155d0f220890d8f790dc56",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "6899c964b8c4a2cf3128"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53246b5aafd59634d424/workflow.json b/records/sd1/sweep/53246b5aafd59634d424/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..fcf5d424d37d91d838118baca94bf23c59a600bc
--- /dev/null
+++ b/records/sd1/sweep/53246b5aafd59634d424/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53246b5aafd59634d424)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.0.in_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5333023d583880b0e064/measurements.json b/records/sd1/sweep/5333023d583880b0e064/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..27670ef14d3dc7c50f3e44dcc706d18df39dd907
--- /dev/null
+++ b/records/sd1/sweep/5333023d583880b0e064/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5333023d583880b0e064",
+ "values": {
+  "cosine_distance": {
+   "value": 0.188368,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.27946,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.164982,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.040567,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.312,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00708,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 28.724,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.976,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0393,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5333023d583880b0e064/record.json b/records/sd1/sweep/5333023d583880b0e064/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3bbd67763362d43eaa698578f87aee060b54c738
--- /dev/null
+++ b/records/sd1/sweep/5333023d583880b0e064/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5333023d583880b0e064",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:39:46Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.1.transformer_blocks.0.attn1.to_out.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "in.hi",
+   "kind": "self_attn",
+   "module_type": "Dropout",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/be89a3f4d9543fc8ea230e930e18756c93519293b4cc225b4a68d3461a4ad87a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "be89a3f4d9543fc8ea230e930e18756c93519293b4cc225b4a68d3461a4ad87a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "617e8f86cef2c1854bf7"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5333023d583880b0e064/workflow.json b/records/sd1/sweep/5333023d583880b0e064/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..255da6edf1245f67cd949f6bc3c74929ffc1dc13
--- /dev/null
+++ b/records/sd1/sweep/5333023d583880b0e064/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5333023d583880b0e064)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.1.transformer_blocks.0.attn1.to_out.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5340a4a8c7bf40946b16/measurements.json b/records/sd1/sweep/5340a4a8c7bf40946b16/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..bc25f298f255256d60954cddfc0cb65be322f722
--- /dev/null
+++ b/records/sd1/sweep/5340a4a8c7bf40946b16/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5340a4a8c7bf40946b16",
+ "values": {
+  "cosine_distance": {
+   "value": 0.395642,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.497361,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.193445,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.1132,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 38.655,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00102,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 44.72,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.994,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4306,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5340a4a8c7bf40946b16/record.json b/records/sd1/sweep/5340a4a8c7bf40946b16/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..beab1790a13c516eb9e93ac31d42b0cb9e3d185d
--- /dev/null
+++ b/records/sd1/sweep/5340a4a8c7bf40946b16/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5340a4a8c7bf40946b16",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:39:39Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.0.out_layers.2",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.hi",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/a3eade891d9ecae78e35690daadf06ad9ad9ec9e65917ac1899b9a16cdd39084.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "a3eade891d9ecae78e35690daadf06ad9ad9ec9e65917ac1899b9a16cdd39084",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "61c8de29ce77e7f213b9"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5340a4a8c7bf40946b16/workflow.json b/records/sd1/sweep/5340a4a8c7bf40946b16/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c6703e37bd281bb4380fd950cf8c5bcc0378be1d
--- /dev/null
+++ b/records/sd1/sweep/5340a4a8c7bf40946b16/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5340a4a8c7bf40946b16)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.0.out_layers.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53507b95f324e9aa172b/measurements.json b/records/sd1/sweep/53507b95f324e9aa172b/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..7a9bcb22241b10208a991ddcde429bde9f2494cd
--- /dev/null
+++ b/records/sd1/sweep/53507b95f324e9aa172b/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "53507b95f324e9aa172b",
+ "values": {
+  "std": {
+   "value": 64.167,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00973,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 66.702,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.60687,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.520989,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.096446,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.089036,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0232,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.138,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53507b95f324e9aa172b/record.json b/records/sd1/sweep/53507b95f324e9aa172b/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..07d2f27ae5b625e07047b7c9dee609837211c52a
--- /dev/null
+++ b/records/sd1/sweep/53507b95f324e9aa172b/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "53507b95f324e9aa172b",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T21:50:51Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180.0
+   },
+   "group": "out.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "half",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.mid",
+   "rotate",
+   180.0,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "184e72894976bdbf3821"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53507b95f324e9aa172b/workflow.json b/records/sd1/sweep/53507b95f324e9aa172b/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f8ebbda700d95ddca8494b362ded82085d45682e
--- /dev/null
+++ b/records/sd1/sweep/53507b95f324e9aa172b/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53507b95f324e9aa172b)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535396098fabbd7ff2ba/measurements.json b/records/sd1/sweep/535396098fabbd7ff2ba/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..ee6e9fae777a6a4ec243ed051697d6654deb1621
--- /dev/null
+++ b/records/sd1/sweep/535396098fabbd7ff2ba/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "535396098fabbd7ff2ba",
+ "values": {
+  "cosine_distance": {
+   "value": 0.209332,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.351916,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.370955,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.276148,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 48.976,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00284,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.432,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9876,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5687,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535396098fabbd7ff2ba/record.json b/records/sd1/sweep/535396098fabbd7ff2ba/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b46ce109775a98882cbc4fa078346a524d0bf4c6
--- /dev/null
+++ b/records/sd1/sweep/535396098fabbd7ff2ba/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "535396098fabbd7ff2ba",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:19Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.out_layers.3",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.5
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/b6894bfe5217dde103acc3550d0f647165afaf307d09a24230251f6bfbb3cd3b.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "b6894bfe5217dde103acc3550d0f647165afaf307d09a24230251f6bfbb3cd3b",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "7245344eaf20ce80010d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535396098fabbd7ff2ba/workflow.json b/records/sd1/sweep/535396098fabbd7ff2ba/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..d169bda282b7397d1c49bda3a9f9476c4ce82e52
--- /dev/null
+++ b/records/sd1/sweep/535396098fabbd7ff2ba/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 535396098fabbd7ff2ba)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.out_layers.3\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535e5512c2337e79b1b6/interpretations.jsonl b/records/sd1/sweep/535e5512c2337e79b1b6/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..2a8fe93e454d36b72613d3ef12eb37b5f1cf66e0
--- /dev/null
+++ b/records/sd1/sweep/535e5512c2337e79b1b6/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "535e5512c2337e79b1b6"}, "kind": "caption", "value": "Black-and-white portrait of a front-facing young woman with slicked-back hair, dark top.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:28:44Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "18b9c602b9afcb65"}
+{"target": {"record": "535e5512c2337e79b1b6"}, "kind": "change", "value": "The image turns monochrome and crisp, with a slightly different, more frontal person with sleeker hair, against a plain grey backdrop.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:28:44Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "c1d7d3ff167d74b8"}
+{"target": {"record": "535e5512c2337e79b1b6"}, "kind": "keywords", "value": ["black and white", "monochrome portrait", "frontal pose", "identity shift", "slicked hair"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:28:44Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ed38acb84f02fc4a"}
+{"target": {"record": "535e5512c2337e79b1b6"}, "kind": "effect_tags", "value": ["desaturated", "subject_kept", "style_change", "high_contrast"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:28:44Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "60264f2c1d53c029"}
diff --git a/records/sd1/sweep/535e5512c2337e79b1b6/measurements.json b/records/sd1/sweep/535e5512c2337e79b1b6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..053f60acebef908a2c19ef98256610bd6b970bec
--- /dev/null
+++ b/records/sd1/sweep/535e5512c2337e79b1b6/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "535e5512c2337e79b1b6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.187296,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.188278,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.237407,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.168087,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 59.843,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00408,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 18.561,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9348,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.6063,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535e5512c2337e79b1b6/record.json b/records/sd1/sweep/535e5512c2337e79b1b6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..856f64dbe32d9b84e720bdf4f72d934ab5ac6d70
--- /dev/null
+++ b/records/sd1/sweep/535e5512c2337e79b1b6/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "535e5512c2337e79b1b6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:20Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.4.0.emb_layers.1",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Linear",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/31d71295c2f66b18dfb0e11dff3e30503810a411f70d24148f303098c4f6da72.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "31d71295c2f66b18dfb0e11dff3e30503810a411f70d24148f303098c4f6da72",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "4df996cc761cba72512d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535e5512c2337e79b1b6/workflow.json b/records/sd1/sweep/535e5512c2337e79b1b6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..0acb4054841c02c9ee3f3756fa6c9f812e9c5fc3
--- /dev/null
+++ b/records/sd1/sweep/535e5512c2337e79b1b6/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 535e5512c2337e79b1b6)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.4.0.emb_layers.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535f4d9071f451ed63fe/measurements.json b/records/sd1/sweep/535f4d9071f451ed63fe/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..562e070c6c956e25af49a58e02a4991962cfb36a
--- /dev/null
+++ b/records/sd1/sweep/535f4d9071f451ed63fe/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "535f4d9071f451ed63fe",
+ "values": {
+  "cosine_distance": {
+   "value": 0.004343,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.013289,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.004157,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.001678,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 59.745,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00571,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 3.4,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0027,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1187,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535f4d9071f451ed63fe/record.json b/records/sd1/sweep/535f4d9071f451ed63fe/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..679709e3f411f181cbaf046771cd03c2f8177d0f
--- /dev/null
+++ b/records/sd1/sweep/535f4d9071f451ed63fe/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "535f4d9071f451ed63fe",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:42:41Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.10.0.in_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/55b8f7d41de6c87b03b309a6a13e7e7c796f8cf0173681efcae15567d5d37507.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "55b8f7d41de6c87b03b309a6a13e7e7c796f8cf0173681efcae15567d5d37507",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6fd28309eaa00096f74f"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535f4d9071f451ed63fe/workflow.json b/records/sd1/sweep/535f4d9071f451ed63fe/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..8e178f727e5e0438578b412980144abbd013cb9d
--- /dev/null
+++ b/records/sd1/sweep/535f4d9071f451ed63fe/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 535f4d9071f451ed63fe)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.10.0.in_layers.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535f97dfb6e59549383d/measurements.json b/records/sd1/sweep/535f97dfb6e59549383d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9e81344988766b2b3759b63a421dc57c703764a
--- /dev/null
+++ b/records/sd1/sweep/535f97dfb6e59549383d/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "535f97dfb6e59549383d",
+ "values": {
+  "std": {
+   "value": 35.434,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01994,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 43.795,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.355269,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.563583,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.19545,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.081649,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.8769,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7253,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535f97dfb6e59549383d/record.json b/records/sd1/sweep/535f97dfb6e59549383d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..8aed770647438771003f8f9404b29d001b9d0a57
--- /dev/null
+++ b/records/sd1/sweep/535f97dfb6e59549383d/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "535f97dfb6e59549383d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:14:58Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": -0.8
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "mid"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "add_scalar",
+   -0.8,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": -0.8,
+    "applied": -0.8,
+    "reason": "outside safe range [-0.5, 1.5] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6c43a13cfb65c9c310af"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/535f97dfb6e59549383d/workflow.json b/records/sd1/sweep/535f97dfb6e59549383d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..917f1d8ae74fc7b89ae448e3fd21894f9bad5c95
--- /dev/null
+++ b/records/sd1/sweep/535f97dfb6e59549383d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 535f97dfb6e59549383d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": -0.8\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/536a4bb966be72d279d3/measurements.json b/records/sd1/sweep/536a4bb966be72d279d3/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..57a4bcd1adb665e01616390b08a1dbae6d0a8fbe
--- /dev/null
+++ b/records/sd1/sweep/536a4bb966be72d279d3/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "536a4bb966be72d279d3",
+ "values": {
+  "std": {
+   "value": 31.647,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02515,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 43.006,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.455294,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.561059,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.295613,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.227446,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.8972,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.8537,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/536a4bb966be72d279d3/record.json b/records/sd1/sweep/536a4bb966be72d279d3/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a81073ca715a28e0ed0e28da32d860fdda48550e
--- /dev/null
+++ b/records/sd1/sweep/536a4bb966be72d279d3/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "536a4bb966be72d279d3",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:19:39Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "2976534844fe80be26b4",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "rotate",
+   180.0,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 180.0,
+    "applied": 180.0,
+    "reason": "outside safe range [5.0, 45.0] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "abe223b525063922ec98"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/536a4bb966be72d279d3/workflow.json b/records/sd1/sweep/536a4bb966be72d279d3/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..4a26da458deca1b0b5ed1537f2ffcb02e80d7005
--- /dev/null
+++ b/records/sd1/sweep/536a4bb966be72d279d3/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 536a4bb966be72d279d3)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180.0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/536b47182a6069b7de56/measurements.json b/records/sd1/sweep/536b47182a6069b7de56/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c99fc5eb436ce1338187363af2e2b70e81281c63
--- /dev/null
+++ b/records/sd1/sweep/536b47182a6069b7de56/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "536b47182a6069b7de56",
+ "values": {
+  "std": {
+   "value": 27.303,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02379,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 33.465,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.449712,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.549609,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.461122,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.157807,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0338,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0057,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/536b47182a6069b7de56/record.json b/records/sd1/sweep/536b47182a6069b7de56/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..fd95355641655b946fe153478450539746d8cd3c
--- /dev/null
+++ b/records/sd1/sweep/536b47182a6069b7de56/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "536b47182a6069b7de56",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:05:16Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90.0
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "quarter",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "rotate",
+   90.0,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "7c336a292a315c53fc15"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/536b47182a6069b7de56/workflow.json b/records/sd1/sweep/536b47182a6069b7de56/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7f4bec5899081bd87625fd5aa160eac4cd74d768
--- /dev/null
+++ b/records/sd1/sweep/536b47182a6069b7de56/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 536b47182a6069b7de56)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53776565d7d5e68fd912/measurements.json b/records/sd1/sweep/53776565d7d5e68fd912/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2181ee9b4dcec352d50a1a0abfbcbc19c2c68f56
--- /dev/null
+++ b/records/sd1/sweep/53776565d7d5e68fd912/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "53776565d7d5e68fd912",
+ "values": {
+  "std": {
+   "value": 43.279,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01162,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 9.117,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.036278,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.128911,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.037162,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.027064,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0264,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0332,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53776565d7d5e68fd912/record.json b/records/sd1/sweep/53776565d7d5e68fd912/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..c217af15b94ec1adfadf205d6ecdf8aaf2bab5e5
--- /dev/null
+++ b/records/sd1/sweep/53776565d7d5e68fd912/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "53776565d7d5e68fd912",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:48:45Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "mid",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "add_scalar",
+   0.8,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "56d62df494447beffe75"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53776565d7d5e68fd912/workflow.json b/records/sd1/sweep/53776565d7d5e68fd912/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..4ac000a1cdf37f1c040cde5e08b0624db74676d2
--- /dev/null
+++ b/records/sd1/sweep/53776565d7d5e68fd912/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53776565d7d5e68fd912)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/537ca8998c2455b20f7a/measurements.json b/records/sd1/sweep/537ca8998c2455b20f7a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..90d54089f71b94546e667a75e689191b35efd8dd
--- /dev/null
+++ b/records/sd1/sweep/537ca8998c2455b20f7a/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "537ca8998c2455b20f7a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.0,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.0,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": -0.0,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 66.477,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00637,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.0,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0093,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2957,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/537ca8998c2455b20f7a/record.json b/records/sd1/sweep/537ca8998c2455b20f7a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..539ccbdbac5e7dcf3e7821367e519c2c8fb96189
--- /dev/null
+++ b/records/sd1/sweep/537ca8998c2455b20f7a/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "537ca8998c2455b20f7a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:27Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0.h_upd",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/f9bf61154b5832ffe7a806155f676b6edecb1e6c0d74f73d712f936f71a1090a.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "f9bf61154b5832ffe7a806155f676b6edecb1e6c0d74f73d712f936f71a1090a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "8dfb5d985664df51e0f7"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/537ca8998c2455b20f7a/workflow.json b/records/sd1/sweep/537ca8998c2455b20f7a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..bc94a87f9ebb91a3de937e856cf926b17d0fd8ca
--- /dev/null
+++ b/records/sd1/sweep/537ca8998c2455b20f7a/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 537ca8998c2455b20f7a)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0.h_upd\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/537dd7e190c55e137744/measurements.json b/records/sd1/sweep/537dd7e190c55e137744/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..95cdfeb76abe168c0100210bedd4ec74f16b3b5c
--- /dev/null
+++ b/records/sd1/sweep/537dd7e190c55e137744/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "537dd7e190c55e137744",
+ "values": {
+  "std": {
+   "value": 71.145,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00465,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 61.541,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.601958,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.679969,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.917721,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.422496,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.6829,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7345,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/537dd7e190c55e137744/record.json b/records/sd1/sweep/537dd7e190c55e137744/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..74c4138b2a8d0a0856a99ad749d69e100793e628
--- /dev/null
+++ b/records/sd1/sweep/537dd7e190c55e137744/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "537dd7e190c55e137744",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T21:56:03Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30.0
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "slight",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "rotate",
+   30.0,
+   "middle"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 30.0,
+    "applied": 30.0,
+    "reason": "outside safe range [0.0, 5.0] for out.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "46cbad568d4f8cf1828e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/537dd7e190c55e137744/workflow.json b/records/sd1/sweep/537dd7e190c55e137744/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..5af181733c7329226b661ac1cca334e1b1336795
--- /dev/null
+++ b/records/sd1/sweep/537dd7e190c55e137744/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 537dd7e190c55e137744)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/538aac1e4ac5753c07c2/measurements.json b/records/sd1/sweep/538aac1e4ac5753c07c2/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..eadab893696ebf8acd4401009b026109012f644d
--- /dev/null
+++ b/records/sd1/sweep/538aac1e4ac5753c07c2/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "538aac1e4ac5753c07c2",
+ "values": {
+  "cosine_distance": {
+   "value": 0.00841,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.016857,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.004163,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.002008,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 59.478,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00531,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 3.946,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0096,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1523,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/538aac1e4ac5753c07c2/record.json b/records/sd1/sweep/538aac1e4ac5753c07c2/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..bdefea58a0447ac506fbd7e918582691d81ff17a
--- /dev/null
+++ b/records/sd1/sweep/538aac1e4ac5753c07c2/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "538aac1e4ac5753c07c2",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:50Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.1.transformer_blocks.0.attn2.to_k",
+   "op": "multiply",
+   "args": {
+    "scalar": 1.5
+   },
+   "group": "out.hi",
+   "kind": "cross_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/755e63356a3ad96c428c091640323ecc6f0fd3820d3251e927470405f0e79511.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "755e63356a3ad96c428c091640323ecc6f0fd3820d3251e927470405f0e79511",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "94c985f971138ff90f29"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/538aac1e4ac5753c07c2/workflow.json b/records/sd1/sweep/538aac1e4ac5753c07c2/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..38bd3deb3a41e4a71b199c4014b139308ed5a050
--- /dev/null
+++ b/records/sd1/sweep/538aac1e4ac5753c07c2/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 538aac1e4ac5753c07c2)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.1.transformer_blocks.0.attn2.to_k\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/538ac800e668cd832ad6/measurements.json b/records/sd1/sweep/538ac800e668cd832ad6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0cc67a0c088660792867919655c85bb1fdc290a3
--- /dev/null
+++ b/records/sd1/sweep/538ac800e668cd832ad6/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "538ac800e668cd832ad6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.231324,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.434767,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.513148,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.283239,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 45.832,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00407,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 27.482,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9809,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5139,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/538ac800e668cd832ad6/record.json b/records/sd1/sweep/538ac800e668cd832ad6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a98ae74aeb680041d51777b1904cd6a6f348da03
--- /dev/null
+++ b/records/sd1/sweep/538ac800e668cd832ad6/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "538ac800e668cd832ad6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:08Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.in_layers.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/9aaa052f1a25e8db3456c1c5e24aba0bcd8bc588e124adce756c460198541eda.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "9aaa052f1a25e8db3456c1c5e24aba0bcd8bc588e124adce756c460198541eda",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "948d6a8d8a1ca2be3559"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/538ac800e668cd832ad6/workflow.json b/records/sd1/sweep/538ac800e668cd832ad6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ad0cbe71275c43e7f4a5e022832d2847be64ef34
--- /dev/null
+++ b/records/sd1/sweep/538ac800e668cd832ad6/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 538ac800e668cd832ad6)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.in_layers.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/539e0d36ea9f56e8cb85/measurements.json b/records/sd1/sweep/539e0d36ea9f56e8cb85/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..300a498d3cca57fc63bffc52660ffdf5b9ab0725
--- /dev/null
+++ b/records/sd1/sweep/539e0d36ea9f56e8cb85/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "539e0d36ea9f56e8cb85",
+ "values": {
+  "std": {
+   "value": 51.623,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00919,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 41.774,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.469629,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.566249,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.381675,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.13557,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0042,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1422,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/539e0d36ea9f56e8cb85/record.json b/records/sd1/sweep/539e0d36ea9f56e8cb85/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..2275a87391575cb925edcaf554bc424d064e0743
--- /dev/null
+++ b/records/sd1/sweep/539e0d36ea9f56e8cb85/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "539e0d36ea9f56e8cb85",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:46:28Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "out.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "grow",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.mid",
+   "scale",
+   1.4,
+   "early"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 1.4,
+    "applied": 1.4,
+    "reason": "outside safe range [0.7, 1.3] for out.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e75a4636ecaa4fc33852"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/539e0d36ea9f56e8cb85/workflow.json b/records/sd1/sweep/539e0d36ea9f56e8cb85/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..04adb0ecb2d545f54cfebdb751256ae086bbd558
--- /dev/null
+++ b/records/sd1/sweep/539e0d36ea9f56e8cb85/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 539e0d36ea9f56e8cb85)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/539f09ccfb92b29b4fad/measurements.json b/records/sd1/sweep/539f09ccfb92b29b4fad/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..88907d78bbc88f903c1153b401ee192f675f7236
--- /dev/null
+++ b/records/sd1/sweep/539f09ccfb92b29b4fad/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "539f09ccfb92b29b4fad",
+ "values": {
+  "cosine_distance": {
+   "value": 0.197471,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.285749,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.14025,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.056738,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 45.117,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00523,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.892,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9277,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0777,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/539f09ccfb92b29b4fad/record.json b/records/sd1/sweep/539f09ccfb92b29b4fad/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..229c4527b6f964f2bd77a93275d4520a082cb376
--- /dev/null
+++ b/records/sd1/sweep/539f09ccfb92b29b4fad/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "539f09ccfb92b29b4fad",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:30Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0.out_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/5aad1533ab9b1f2334902704b3385a73459f05b2918ceafb1883ef611d3ec27a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "5aad1533ab9b1f2334902704b3385a73459f05b2918ceafb1883ef611d3ec27a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "879709100bfacaa45af6"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/539f09ccfb92b29b4fad/workflow.json b/records/sd1/sweep/539f09ccfb92b29b4fad/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..09e03dc45115040436597b888997ec44d0d0749e
--- /dev/null
+++ b/records/sd1/sweep/539f09ccfb92b29b4fad/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 539f09ccfb92b29b4fad)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0.out_layers.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53bac38087ca9a7249f7/measurements.json b/records/sd1/sweep/53bac38087ca9a7249f7/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6c0720dd4b7452f7943690b832c248b8692ad703
--- /dev/null
+++ b/records/sd1/sweep/53bac38087ca9a7249f7/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "53bac38087ca9a7249f7",
+ "values": {
+  "cosine_distance": {
+   "value": 0.024963,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.040287,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.029109,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.00805,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.114,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00539,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 6.336,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9861,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2132,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53bac38087ca9a7249f7/record.json b/records/sd1/sweep/53bac38087ca9a7249f7/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..87cfe41659856c78ef7ad0a64ea3f5b65283d6a4
--- /dev/null
+++ b/records/sd1/sweep/53bac38087ca9a7249f7/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "53bac38087ca9a7249f7",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:42Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.1.norm",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "out.hi",
+   "kind": "norm",
+   "module_type": "GroupNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/8c597492dfde2950da8dad65c7ec1afe67124cea4f49b31a1d23a46586c6a28e.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "8c597492dfde2950da8dad65c7ec1afe67124cea4f49b31a1d23a46586c6a28e",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e49b2de7b359af688e03"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53bac38087ca9a7249f7/workflow.json b/records/sd1/sweep/53bac38087ca9a7249f7/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..434d3535a81facb2911cfe5a43402ff1339126c0
--- /dev/null
+++ b/records/sd1/sweep/53bac38087ca9a7249f7/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53bac38087ca9a7249f7)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.1.norm\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53bdd990e60b127b4079/measurements.json b/records/sd1/sweep/53bdd990e60b127b4079/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..9986f1a5650e747cd6f25f308a71bd96eecf9e00
--- /dev/null
+++ b/records/sd1/sweep/53bdd990e60b127b4079/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "53bdd990e60b127b4079",
+ "values": {
+  "std": {
+   "value": 47.588,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00166,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": true,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": [
+    "clip_degenerate"
+   ]
+  },
+  "mae_vs_baseline": {
+   "value": 18.008,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.161904,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.545058,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.765397,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.33572,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.7551,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.902,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53bdd990e60b127b4079/record.json b/records/sd1/sweep/53bdd990e60b127b4079/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5d823f508783e73a3b612cb1ec04f0e1464ed690
--- /dev/null
+++ b/records/sd1/sweep/53bdd990e60b127b4079/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "53bdd990e60b127b4079",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:48:26Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30.0
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "slight",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "rotate",
+   30.0,
+   "late"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 30.0,
+    "applied": 30.0,
+    "reason": "outside safe range [0.0, 5.0] for out.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6b82f3e6e57765793083"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53bdd990e60b127b4079/workflow.json b/records/sd1/sweep/53bdd990e60b127b4079/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a652c8128ac48ade9a06ff76d9bb8033c4370547
--- /dev/null
+++ b/records/sd1/sweep/53bdd990e60b127b4079/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53bdd990e60b127b4079)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30.0\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53c1c161aab74865ee90/measurements.json b/records/sd1/sweep/53c1c161aab74865ee90/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..210f5bd8f8f5cffc1f22dabe360a116fadc738f4
--- /dev/null
+++ b/records/sd1/sweep/53c1c161aab74865ee90/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "53c1c161aab74865ee90",
+ "values": {
+  "std": {
+   "value": 70.301,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00325,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 66.972,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.611282,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.75159,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.99075,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.379564,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.5877,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7992,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53c1c161aab74865ee90/record.json b/records/sd1/sweep/53c1c161aab74865ee90/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a7daa1d338e992b8f10d868ca2d1b8b632ee184e
--- /dev/null
+++ b/records/sd1/sweep/53c1c161aab74865ee90/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "53c1c161aab74865ee90",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:11:06Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "scale",
+   1.4,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 1.4,
+    "applied": 1.4,
+    "reason": "outside safe range [0.95, 1.05] for out.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "0052711693b1d7481087"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53c1c161aab74865ee90/workflow.json b/records/sd1/sweep/53c1c161aab74865ee90/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..30ba70b8269181f307db28c4e2819ecee6648b43
--- /dev/null
+++ b/records/sd1/sweep/53c1c161aab74865ee90/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53c1c161aab74865ee90)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53c2e3e73ab0b0af3f46/measurements.json b/records/sd1/sweep/53c2e3e73ab0b0af3f46/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..9c9d17b3ed68d9ccf4e23bee240cb22223b3e29d
--- /dev/null
+++ b/records/sd1/sweep/53c2e3e73ab0b0af3f46/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "53c2e3e73ab0b0af3f46",
+ "values": {
+  "std": {
+   "value": 45.724,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01446,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 37.583,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.412737,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.540457,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.144352,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.154691,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0093,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1401,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53c2e3e73ab0b0af3f46/record.json b/records/sd1/sweep/53c2e3e73ab0b0af3f46/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..754da91899a6103b546237260a2a034d9553bc56
--- /dev/null
+++ b/records/sd1/sweep/53c2e3e73ab0b0af3f46/record.json
@@ -0,0 +1,77 @@
+{
+ "id": "53c2e3e73ab0b0af3f46",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:38:12Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "portrait of an old fisherman, analog photograph",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "27bbff3e15d029a6"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.7
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "shrink",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "c244b02b7723b34ac353",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "scale",
+   0.7,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53c2e3e73ab0b0af3f46/workflow.json b/records/sd1/sweep/53c2e3e73ab0b0af3f46/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..d06012fa0175f197d1017a68d717c949596da9ec
--- /dev/null
+++ b/records/sd1/sweep/53c2e3e73ab0b0af3f46/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53c2e3e73ab0b0af3f46)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.7\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "portrait of an old fisherman, analog photograph"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53d90b7f4267d56cebc5/measurements.json b/records/sd1/sweep/53d90b7f4267d56cebc5/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..bd19c989d4aa6e575e68dc9aba1330d1f953a419
--- /dev/null
+++ b/records/sd1/sweep/53d90b7f4267d56cebc5/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "53d90b7f4267d56cebc5",
+ "values": {
+  "cosine_distance": {
+   "value": 0.078755,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.126669,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.041387,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.029424,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 57.863,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00749,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 13.574,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9631,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0649,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53d90b7f4267d56cebc5/record.json b/records/sd1/sweep/53d90b7f4267d56cebc5/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..e661097e76dda06c9c5a123db9721ae1abd8ec7e
--- /dev/null
+++ b/records/sd1/sweep/53d90b7f4267d56cebc5/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "53d90b7f4267d56cebc5",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:42:14Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.7.1.transformer_blocks.0.norm2",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.lo",
+   "kind": "norm",
+   "module_type": "LayerNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/331dfaa354a9c4f7e7b6ceafea549a75a3492b0b5eacacda6514f6079f27944e.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "331dfaa354a9c4f7e7b6ceafea549a75a3492b0b5eacacda6514f6079f27944e",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.LayerNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "289c98fa385530701760"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53d90b7f4267d56cebc5/workflow.json b/records/sd1/sweep/53d90b7f4267d56cebc5/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..418065ef4ab31d3a91e1fb6abde1311b061a5945
--- /dev/null
+++ b/records/sd1/sweep/53d90b7f4267d56cebc5/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53d90b7f4267d56cebc5)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.7.1.transformer_blocks.0.norm2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53ddaade083d4a9270e8/interpretations.jsonl b/records/sd1/sweep/53ddaade083d4a9270e8/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..e820231cb297cb52991690a257c0defa9fbbc7ca
--- /dev/null
+++ b/records/sd1/sweep/53ddaade083d4a9270e8/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "53ddaade083d4a9270e8"}, "kind": "caption", "value": "Blurry glowing orange flowers and vase framed by a smoky dark brown border.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:45:02Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "c767bd52307febc7"}
+{"target": {"record": "53ddaade083d4a9270e8"}, "kind": "change", "value": "The crisp still life of glass jars and fruit melts into a hazy, glowing vase-and-flowers blob surrounded by a dark smudged vignette frame.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:45:02Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "69195548e2f4c32a"}
+{"target": {"record": "53ddaade083d4a9270e8"}, "kind": "keywords", "value": ["smoky frame", "blurred still life", "glowing vase", "vignette border", "melted fruit"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:45:02Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "e7517b34007be643"}
+{"target": {"record": "53ddaade083d4a9270e8"}, "kind": "effect_tags", "value": ["framing", "blur", "painterly", "distorted", "subject_lost"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:45:02Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "f75d1471a2642877"}
diff --git a/records/sd1/sweep/53ddaade083d4a9270e8/measurements.json b/records/sd1/sweep/53ddaade083d4a9270e8/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e302d3e2a480d0603d62103c2f16244d9adf2143
--- /dev/null
+++ b/records/sd1/sweep/53ddaade083d4a9270e8/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "53ddaade083d4a9270e8",
+ "values": {
+  "std": {
+   "value": 52.799,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0034,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 46.669,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.581712,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.667506,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.668814,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.313905,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.8071,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1607,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53ddaade083d4a9270e8/record.json b/records/sd1/sweep/53ddaade083d4a9270e8/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..21801fe076afed744d7de2b1ace8c9be27eb21f9
--- /dev/null
+++ b/records/sd1/sweep/53ddaade083d4a9270e8/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "53ddaade083d4a9270e8",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:31:37Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.7
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "scale",
+   0.7,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.7,
+    "applied": 0.7,
+    "reason": "outside safe range [0.95, 1.05] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a1f06c37d068d0b09cf0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53ddaade083d4a9270e8/workflow.json b/records/sd1/sweep/53ddaade083d4a9270e8/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..723dbfcd5d01acf212064ff570ca2989b35cc1d7
--- /dev/null
+++ b/records/sd1/sweep/53ddaade083d4a9270e8/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53ddaade083d4a9270e8)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.7\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53f6b9e2b5089320452c/measurements.json b/records/sd1/sweep/53f6b9e2b5089320452c/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..ff17b4da5d41f39bf86ed9e5f53de0a15c0b84f4
--- /dev/null
+++ b/records/sd1/sweep/53f6b9e2b5089320452c/measurements.json
@@ -0,0 +1,58 @@
+{
+ "record": "53f6b9e2b5089320452c",
+ "values": {
+  "cosine_distance": {
+   "value": 0.434936,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.623012,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.608532,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.404669,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 118.503,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00184,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "clipped",
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 76.047,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9939,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.8301,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53f6b9e2b5089320452c/record.json b/records/sd1/sweep/53f6b9e2b5089320452c/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..db57b773b51bcf336a66da414fb2689814e52b70
--- /dev/null
+++ b/records/sd1/sweep/53f6b9e2b5089320452c/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "53f6b9e2b5089320452c",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:38Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.1.transformer_blocks.0.attn1.to_k",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "out.mid",
+   "kind": "self_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/5f54d6d8789704b94c4ea28953911d6302e79848c9076e074239a8c016d0e407.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "5f54d6d8789704b94c4ea28953911d6302e79848c9076e074239a8c016d0e407",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "89b16ef7eddbbbd85f5b"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53f6b9e2b5089320452c/workflow.json b/records/sd1/sweep/53f6b9e2b5089320452c/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..e267dbc95ef2f64002d5c8167f66838a27f2061f
--- /dev/null
+++ b/records/sd1/sweep/53f6b9e2b5089320452c/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53f6b9e2b5089320452c)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.1.transformer_blocks.0.attn1.to_k\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53fd706662d034e6c84c/measurements.json b/records/sd1/sweep/53fd706662d034e6c84c/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cf2307187bf2f292a71d2a003bf37d4e859a8ca6
--- /dev/null
+++ b/records/sd1/sweep/53fd706662d034e6c84c/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "53fd706662d034e6c84c",
+ "values": {
+  "cosine_distance": {
+   "value": 0.125695,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.144254,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.064129,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.083099,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 62.523,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0061,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 17.744,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9902,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3055,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53fd706662d034e6c84c/record.json b/records/sd1/sweep/53fd706662d034e6c84c/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7e3bc88ff1c689393ba2ce8a3a4efb8fa0560490
--- /dev/null
+++ b/records/sd1/sweep/53fd706662d034e6c84c/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "53fd706662d034e6c84c",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:28Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "in.lo",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/9cfe9203fe089d7e590c8d8e3dbfdc39344d4c4cd6298745ccc8ded3dcb2c499.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "9cfe9203fe089d7e590c8d8e3dbfdc39344d4c4cd6298745ccc8ded3dcb2c499",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "dd2dcd4cf37c130115c0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/53fd706662d034e6c84c/workflow.json b/records/sd1/sweep/53fd706662d034e6c84c/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..74b4d3f9f015df594056477611a3354b431d0fcc
--- /dev/null
+++ b/records/sd1/sweep/53fd706662d034e6c84c/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 53fd706662d034e6c84c)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/541214a812d93b383b7b/measurements.json b/records/sd1/sweep/541214a812d93b383b7b/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..eb892ba85a1018b4704f8602b9286343f3fa22e1
--- /dev/null
+++ b/records/sd1/sweep/541214a812d93b383b7b/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "541214a812d93b383b7b",
+ "values": {
+  "std": {
+   "value": 38.431,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01821,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 18.429,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.096372,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.186402,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.042021,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.029887,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0624,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0192,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/541214a812d93b383b7b/record.json b/records/sd1/sweep/541214a812d93b383b7b/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..cfe188cfa28be6cad4e5c2bbaca6849a7351df53
--- /dev/null
+++ b/records/sd1/sweep/541214a812d93b383b7b/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "541214a812d93b383b7b",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:00:30Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "mid",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "add_scalar",
+   0.8,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e3218fe7b5a27f1a858e"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/541214a812d93b383b7b/workflow.json b/records/sd1/sweep/541214a812d93b383b7b/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..9666b40005e2c307775265c3af3ad99f85ab2269
--- /dev/null
+++ b/records/sd1/sweep/541214a812d93b383b7b/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 541214a812d93b383b7b)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5414e737c1e2f7e35ca8/measurements.json b/records/sd1/sweep/5414e737c1e2f7e35ca8/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f70f414ccf945c83822772ffc38afcb6c78fc891
--- /dev/null
+++ b/records/sd1/sweep/5414e737c1e2f7e35ca8/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5414e737c1e2f7e35ca8",
+ "values": {
+  "cosine_distance": {
+   "value": 0.063257,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.102131,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.059534,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.01783,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 62.249,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00494,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 12.284,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9848,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1512,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5414e737c1e2f7e35ca8/record.json b/records/sd1/sweep/5414e737c1e2f7e35ca8/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..8a04856e2f00817f60889a0a359fd2ad1aff28ac
--- /dev/null
+++ b/records/sd1/sweep/5414e737c1e2f7e35ca8/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5414e737c1e2f7e35ca8",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:49:04Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "middle_block.2.in_layers.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "mid",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/53bf315f8b30541bcafd216cf11659e77701a958ae576dfcaa8b0d4a3062161c.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "53bf315f8b30541bcafd216cf11659e77701a958ae576dfcaa8b0d4a3062161c",
+  "type_path": "UNetModel.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "490a8a3ffd633f9cafa4"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5414e737c1e2f7e35ca8/workflow.json b/records/sd1/sweep/5414e737c1e2f7e35ca8/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c38cbac6043680027bc5b56bed68a323015b4d90
--- /dev/null
+++ b/records/sd1/sweep/5414e737c1e2f7e35ca8/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5414e737c1e2f7e35ca8)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.2.in_layers.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/542874016bcf6b4e5007/measurements.json b/records/sd1/sweep/542874016bcf6b4e5007/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..458d4e33c20e22eb7386f123097ab8b9e26a0fa5
--- /dev/null
+++ b/records/sd1/sweep/542874016bcf6b4e5007/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "542874016bcf6b4e5007",
+ "values": {
+  "std": {
+   "value": 32.143,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02459,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 23.283,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.207944,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.333507,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.082822,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.04558,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0273,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1514,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/542874016bcf6b4e5007/record.json b/records/sd1/sweep/542874016bcf6b4e5007/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..20c935084f9e1e9f821b310057afa5f90e56b954
--- /dev/null
+++ b/records/sd1/sweep/542874016bcf6b4e5007/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "542874016bcf6b4e5007",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:53:12Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 2.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "boost",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "multiply",
+   2.0,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6446dfac6bb48e68e410"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/542874016bcf6b4e5007/workflow.json b/records/sd1/sweep/542874016bcf6b4e5007/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c3e815f067e2edccf7690dfaeccf5d6a5c7c3ebf
--- /dev/null
+++ b/records/sd1/sweep/542874016bcf6b4e5007/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 542874016bcf6b4e5007)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2.0\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54341f31b89e9b51d619/measurements.json b/records/sd1/sweep/54341f31b89e9b51d619/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c50e027cd1451212e3bd1415367224ea1c6add7
--- /dev/null
+++ b/records/sd1/sweep/54341f31b89e9b51d619/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "54341f31b89e9b51d619",
+ "values": {
+  "std": {
+   "value": 41.393,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01644,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 32.065,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.315623,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.397276,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.098295,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.066545,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0095,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0318,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54341f31b89e9b51d619/record.json b/records/sd1/sweep/54341f31b89e9b51d619/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..241fabfa34c430a0aefee6180470eed008a29ccc
--- /dev/null
+++ b/records/sd1/sweep/54341f31b89e9b51d619/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "54341f31b89e9b51d619",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:51:54Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "grow",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "scale",
+   1.4,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "5a9626f10ccc4a0fd3a8"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54341f31b89e9b51d619/workflow.json b/records/sd1/sweep/54341f31b89e9b51d619/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7cf9acdafbcbf2643664c413ba540d728bc2ecb6
--- /dev/null
+++ b/records/sd1/sweep/54341f31b89e9b51d619/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54341f31b89e9b51d619)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54394c364269fef02923/measurements.json b/records/sd1/sweep/54394c364269fef02923/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..7e34ce16fe375719d75f58c12287394d07433f4d
--- /dev/null
+++ b/records/sd1/sweep/54394c364269fef02923/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "54394c364269fef02923",
+ "values": {
+  "cosine_distance": {
+   "value": 0.152755,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.182136,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.203001,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.244937,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.409,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00508,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 19.144,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9557,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5328,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54394c364269fef02923/record.json b/records/sd1/sweep/54394c364269fef02923/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..0ffb00a8cdac7d57ef582200b727225391f9c6a2
--- /dev/null
+++ b/records/sd1/sweep/54394c364269fef02923/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "54394c364269fef02923",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:50:51Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.2.0.skip_connection",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "out.lo",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/9683ef59e4aa2805dd4851f6989f725ff46811f0326fa104b25d3dde7bef4875.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "9683ef59e4aa2805dd4851f6989f725ff46811f0326fa104b25d3dde7bef4875",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "304d78bbb8b8b54bde16"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54394c364269fef02923/workflow.json b/records/sd1/sweep/54394c364269fef02923/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..fedc4792b2a726ff5a4e7f36d44e82de580a9339
--- /dev/null
+++ b/records/sd1/sweep/54394c364269fef02923/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54394c364269fef02923)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.2.0.skip_connection\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/543d86af58d2c2a842f0/measurements.json b/records/sd1/sweep/543d86af58d2c2a842f0/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f2bc2d903d6bced06eadfa981f34d74cd4b4c395
--- /dev/null
+++ b/records/sd1/sweep/543d86af58d2c2a842f0/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "543d86af58d2c2a842f0",
+ "values": {
+  "cosine_distance": {
+   "value": 0.190096,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.294068,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.234777,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.181282,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.284,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00281,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.663,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9626,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5003,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/543d86af58d2c2a842f0/record.json b/records/sd1/sweep/543d86af58d2c2a842f0/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ffc2407061c4ef69bcdb25d8e0858634ee3076b
--- /dev/null
+++ b/records/sd1/sweep/543d86af58d2c2a842f0/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "543d86af58d2c2a842f0",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:49Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.0.out_layers.3",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/7d4ba8a75b9c3515174fc609073f2a658afa7c48670013dbc6a6f77e66d749d7.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "7d4ba8a75b9c3515174fc609073f2a658afa7c48670013dbc6a6f77e66d749d7",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "00f26ec386eeb28ddb94"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/543d86af58d2c2a842f0/workflow.json b/records/sd1/sweep/543d86af58d2c2a842f0/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f07decf74ca699a9659c549ecbaa70f6c084f9ed
--- /dev/null
+++ b/records/sd1/sweep/543d86af58d2c2a842f0/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 543d86af58d2c2a842f0)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.0.out_layers.3\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5443814efda111b4d421/measurements.json b/records/sd1/sweep/5443814efda111b4d421/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f6634acc191b24779d710c8ae8754e560e8c4244
--- /dev/null
+++ b/records/sd1/sweep/5443814efda111b4d421/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5443814efda111b4d421",
+ "values": {
+  "cosine_distance": {
+   "value": 0.140086,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.197122,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.053178,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.062274,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 61.07,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00395,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 18.459,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0319,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3815,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5443814efda111b4d421/record.json b/records/sd1/sweep/5443814efda111b4d421/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3cde5f9a341f57cc69806395a221578f47c21d91
--- /dev/null
+++ b/records/sd1/sweep/5443814efda111b4d421/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "5443814efda111b4d421",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:35Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.1.transformer_blocks.0.attn2.to_out.1",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "out.mid",
+   "kind": "cross_attn",
+   "module_type": "Dropout",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/00970626dfe624dac29879633c8a7075121de25deac188f66fc1b31cb43bd8da.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "00970626dfe624dac29879633c8a7075121de25deac188f66fc1b31cb43bd8da",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "2cecbca4a339b53ba899"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5443814efda111b4d421/workflow.json b/records/sd1/sweep/5443814efda111b4d421/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..597550b455c4b25ce75373be790f8d269240bd34
--- /dev/null
+++ b/records/sd1/sweep/5443814efda111b4d421/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5443814efda111b4d421)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.1.transformer_blocks.0.attn2.to_out.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54564fe45f6dc4503f0f/measurements.json b/records/sd1/sweep/54564fe45f6dc4503f0f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..3ca05cf68a936ff70c384aae3ad9fbb0b1933ff2
--- /dev/null
+++ b/records/sd1/sweep/54564fe45f6dc4503f0f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "54564fe45f6dc4503f0f",
+ "values": {
+  "std": {
+   "value": 49.351,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0111,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 18.982,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.24128,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.378895,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.265936,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.113398,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9784,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0508,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54564fe45f6dc4503f0f/record.json b/records/sd1/sweep/54564fe45f6dc4503f0f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..4d641e3703779bb0ed81e9b41d5156681545a069
--- /dev/null
+++ b/records/sd1/sweep/54564fe45f6dc4503f0f/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "54564fe45f6dc4503f0f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:38:43Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "mid",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "27f5baccf7c0d4f1de23",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "add_scalar",
+   0.8,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e624197d5abcb780aecc"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54564fe45f6dc4503f0f/workflow.json b/records/sd1/sweep/54564fe45f6dc4503f0f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..3e4c2372be1f3b5571e586c0e5f208153bd36f9b
--- /dev/null
+++ b/records/sd1/sweep/54564fe45f6dc4503f0f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54564fe45f6dc4503f0f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/545c4a585deeda6e42e2/measurements.json b/records/sd1/sweep/545c4a585deeda6e42e2/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..4adef6bfe0806d9127c7f11ced9afe1ac5321c89
--- /dev/null
+++ b/records/sd1/sweep/545c4a585deeda6e42e2/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "545c4a585deeda6e42e2",
+ "values": {
+  "cosine_distance": {
+   "value": 0.183625,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.383724,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.275794,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.122039,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 51.938,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00467,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 24.115,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0422,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0704,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/545c4a585deeda6e42e2/record.json b/records/sd1/sweep/545c4a585deeda6e42e2/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..a0c344e5c85eacca4a973b3b8a076f07496d3c75
--- /dev/null
+++ b/records/sd1/sweep/545c4a585deeda6e42e2/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "545c4a585deeda6e42e2",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:50:18Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.out_layers.3",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/7b7139db698f3e2323ce1e7865c627e1fa9636984d1903bd90e2fe802cc47eda.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "7b7139db698f3e2323ce1e7865c627e1fa9636984d1903bd90e2fe802cc47eda",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e914a7fca98f7b73942b"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/545c4a585deeda6e42e2/workflow.json b/records/sd1/sweep/545c4a585deeda6e42e2/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..818b42ba1a228c7136b7744a5d3f88fe7858b9a3
--- /dev/null
+++ b/records/sd1/sweep/545c4a585deeda6e42e2/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 545c4a585deeda6e42e2)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.out_layers.3\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5463fa58c7363a2333bc/measurements.json b/records/sd1/sweep/5463fa58c7363a2333bc/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..a8d5d961a44454f5fa08f1c753cc496499b36299
--- /dev/null
+++ b/records/sd1/sweep/5463fa58c7363a2333bc/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5463fa58c7363a2333bc",
+ "values": {
+  "cosine_distance": {
+   "value": 0.06105,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.107419,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.031471,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.013525,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.754,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00828,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 14.782,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9638,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1601,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5463fa58c7363a2333bc/record.json b/records/sd1/sweep/5463fa58c7363a2333bc/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..2dbb108618da0257c7cbd7969a6db3cc6fdd1c7b
--- /dev/null
+++ b/records/sd1/sweep/5463fa58c7363a2333bc/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5463fa58c7363a2333bc",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:03Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.5.0.in_layers.2",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/e15c27b33c0aeb70aedd45593f3206faddda5274b59fbb6f1f73a9950216cfd0.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "e15c27b33c0aeb70aedd45593f3206faddda5274b59fbb6f1f73a9950216cfd0",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "f35d500619147fc7fdd3"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5463fa58c7363a2333bc/workflow.json b/records/sd1/sweep/5463fa58c7363a2333bc/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..06213e832fe6ec65fe5a418f970a0a3166f0f6e3
--- /dev/null
+++ b/records/sd1/sweep/5463fa58c7363a2333bc/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5463fa58c7363a2333bc)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.5.0.in_layers.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/546cf1dfe1b3eee8660f/interpretations.jsonl b/records/sd1/sweep/546cf1dfe1b3eee8660f/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..5a449a0dc524670100ff50a9ad53722ff20b0eb5
--- /dev/null
+++ b/records/sd1/sweep/546cf1dfe1b3eee8660f/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "546cf1dfe1b3eee8660f"}, "kind": "caption", "value": "Massive grey rock spire dominates right side, small lighthouse merged into its peak", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:46:26Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "021c36fbaaecd3ce"}
+{"target": {"record": "546cf1dfe1b3eee8660f"}, "kind": "change", "value": "The lighthouse is absorbed into a huge craggy rock pillar that fills the right side, the beach vanishes and the sunset sky becomes paler and hazier.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:46:26Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "870fa9e7d363d592"}
+{"target": {"record": "546cf1dfe1b3eee8660f"}, "kind": "keywords", "value": ["rock spire", "giant cliff", "absorbed lighthouse", "hazy sunset", "rocky monolith"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:46:26Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "d89a6d12a1b520cb"}
+{"target": {"record": "546cf1dfe1b3eee8660f"}, "kind": "effect_tags", "value": ["new_composition", "subject_lost", "surreal", "washed_out"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:46:26Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "04369b04afddd92e"}
diff --git a/records/sd1/sweep/546cf1dfe1b3eee8660f/measurements.json b/records/sd1/sweep/546cf1dfe1b3eee8660f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..03cf93a19e9befe8c81adf993a31bce54d5dd8d4
--- /dev/null
+++ b/records/sd1/sweep/546cf1dfe1b3eee8660f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "546cf1dfe1b3eee8660f",
+ "values": {
+  "std": {
+   "value": 37.447,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01617,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 33.141,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.365933,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.523971,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.218984,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.196611,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0201,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.024,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/546cf1dfe1b3eee8660f/record.json b/records/sd1/sweep/546cf1dfe1b3eee8660f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..170b5463458592fd5413712775db69b02afb61de
--- /dev/null
+++ b/records/sd1/sweep/546cf1dfe1b3eee8660f/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "546cf1dfe1b3eee8660f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T01:51:18Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.8.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30.0
+   },
+   "group": "in.lo",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "slight",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "a77bd19c2cc87d81e6d0",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.lo",
+   "rotate",
+   30.0,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a884299fbe05a1a28bb6"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/546cf1dfe1b3eee8660f/workflow.json b/records/sd1/sweep/546cf1dfe1b3eee8660f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..8ff033bab897d18da5ac4df9b9af351021664922
--- /dev/null
+++ b/records/sd1/sweep/546cf1dfe1b3eee8660f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 546cf1dfe1b3eee8660f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.8.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54742b32ef2947e0ab1d/measurements.json b/records/sd1/sweep/54742b32ef2947e0ab1d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..356a2e33e3ea415a80f6eac20780db4763e152d9
--- /dev/null
+++ b/records/sd1/sweep/54742b32ef2947e0ab1d/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "54742b32ef2947e0ab1d",
+ "values": {
+  "cosine_distance": {
+   "value": 0.181949,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.180263,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.11455,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.11252,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 52.914,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00361,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 19.724,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9206,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7719,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54742b32ef2947e0ab1d/record.json b/records/sd1/sweep/54742b32ef2947e0ab1d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1340abc315a7b071e39b039426a00ac963850aff
--- /dev/null
+++ b/records/sd1/sweep/54742b32ef2947e0ab1d/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "54742b32ef2947e0ab1d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:04Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.5.1.proj_out",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.lo",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/7ea4354f71948c57af49847213d832145fbc649a9724c1d358e3561d3544d42a.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "7ea4354f71948c57af49847213d832145fbc649a9724c1d358e3561d3544d42a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "be4cd2b039118ecd1f72"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54742b32ef2947e0ab1d/workflow.json b/records/sd1/sweep/54742b32ef2947e0ab1d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..6603dadaf3dcb14659ab8aee29be9c6fb7dd265d
--- /dev/null
+++ b/records/sd1/sweep/54742b32ef2947e0ab1d/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54742b32ef2947e0ab1d)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.5.1.proj_out\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/547577376505b801b678/measurements.json b/records/sd1/sweep/547577376505b801b678/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e43891ff7c655baafe6df68e64fd8798840086f6
--- /dev/null
+++ b/records/sd1/sweep/547577376505b801b678/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "547577376505b801b678",
+ "values": {
+  "cosine_distance": {
+   "value": 0.056292,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.115894,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.035919,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.024846,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 62.909,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00542,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 11.696,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0387,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0907,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/547577376505b801b678/record.json b/records/sd1/sweep/547577376505b801b678/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..486a97d70af187525879b52a030b866435e79d03
--- /dev/null
+++ b/records/sd1/sweep/547577376505b801b678/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "547577376505b801b678",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:46:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.1.proj_in",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.hi",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/f23be1a5b6f24546f8ab195f843a5743dba2fb666ec6b12709ed236a448fe891.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "f23be1a5b6f24546f8ab195f843a5743dba2fb666ec6b12709ed236a448fe891",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "5198f946aecdbe7e4a70"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/547577376505b801b678/workflow.json b/records/sd1/sweep/547577376505b801b678/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..ec067a094e272c781589f7db292906c7cb241980
--- /dev/null
+++ b/records/sd1/sweep/547577376505b801b678/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 547577376505b801b678)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.1.proj_in\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/547af5b98b94a9c116f2/interpretations.jsonl b/records/sd1/sweep/547af5b98b94a9c116f2/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..02d0829933769e4ba1a707213b01f7aa57a71254
--- /dev/null
+++ b/records/sd1/sweep/547af5b98b94a9c116f2/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "547af5b98b94a9c116f2"}, "kind": "caption", "value": "Ghostly, heavily blurred monochrome portrait of a pale woman with dark wavy hair", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:14:23Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "65342f29fc0c8403"}
+{"target": {"record": "547af5b98b94a9c116f2"}, "kind": "change", "value": "The portrait goes nearly monochrome and very blurry: features fade into a ghostly pale face, colour is almost gone, and the background becomes flat grey haze.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:14:23Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3513647ab1866d7a"}
+{"target": {"record": "547af5b98b94a9c116f2"}, "kind": "keywords", "value": ["ghostly", "heavy blur", "monochrome", "faded portrait", "dreamlike"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:14:23Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "81087d9223ab49fb"}
+{"target": {"record": "547af5b98b94a9c116f2"}, "kind": "effect_tags", "value": ["blur", "desaturated", "washed_out", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:14:23Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "49780937e7ffcc55"}
diff --git a/records/sd1/sweep/547af5b98b94a9c116f2/measurements.json b/records/sd1/sweep/547af5b98b94a9c116f2/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..3917d0d180f7f26a41b428959df88e4ec865da25
--- /dev/null
+++ b/records/sd1/sweep/547af5b98b94a9c116f2/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "547af5b98b94a9c116f2",
+ "values": {
+  "cosine_distance": {
+   "value": 0.255687,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.389822,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.123731,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.11239,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 49.655,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00347,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 29.784,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9463,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0753,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/547af5b98b94a9c116f2/record.json b/records/sd1/sweep/547af5b98b94a9c116f2/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3ed114f378b101f17fbafe98883f496d2fa011b9
--- /dev/null
+++ b/records/sd1/sweep/547af5b98b94a9c116f2/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "547af5b98b94a9c116f2",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:40:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.1.proj_out",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.hi",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/c8857dc9ccc8c13969606d3b3c0115c7758fa67ccff51b7dc8e81deeba9a8c20.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "c8857dc9ccc8c13969606d3b3c0115c7758fa67ccff51b7dc8e81deeba9a8c20",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "1f22eb30f2bde5dc762f"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/547af5b98b94a9c116f2/workflow.json b/records/sd1/sweep/547af5b98b94a9c116f2/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..cff63fa2de850fa20acbaa17382ec3f0aae830f5
--- /dev/null
+++ b/records/sd1/sweep/547af5b98b94a9c116f2/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 547af5b98b94a9c116f2)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.1.proj_out\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54875ff04a29c2137df5/interpretations.jsonl b/records/sd1/sweep/54875ff04a29c2137df5/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..a09f00423fed5455499d925419d42095e6e08829
--- /dev/null
+++ b/records/sd1/sweep/54875ff04a29c2137df5/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "54875ff04a29c2137df5"}, "kind": "caption", "value": "Heavily blurred glowing face shape within dark brown and blue haze", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:13:37Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "99537fa3a5f6d68f"}
+{"target": {"record": "54875ff04a29c2137df5"}, "kind": "change", "value": "The portrait becomes a soft, defocused blur in which only a glowing, tilted face-like light area surrounded by dark hair-like shadows remains.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:13:37Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "c04545cccf1494a8"}
+{"target": {"record": "54875ff04a29c2137df5"}, "kind": "keywords", "value": ["defocused portrait", "glowing face", "dreamy blur", "dark haze", "ghostly"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:13:37Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "8f07da9ad658edc0"}
+{"target": {"record": "54875ff04a29c2137df5"}, "kind": "effect_tags", "value": ["blur", "abstract", "distorted", "dark", "subject_lost"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:13:37Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "c3ddc6e776169e9f"}
diff --git a/records/sd1/sweep/54875ff04a29c2137df5/measurements.json b/records/sd1/sweep/54875ff04a29c2137df5/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c8d875c8b9c0b90ab44df6fabde140eaed8dd702
--- /dev/null
+++ b/records/sd1/sweep/54875ff04a29c2137df5/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "54875ff04a29c2137df5",
+ "values": {
+  "cosine_distance": {
+   "value": 0.374581,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.678111,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.88698,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.293799,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 56.977,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00227,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 38.744,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0317,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2883,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54875ff04a29c2137df5/record.json b/records/sd1/sweep/54875ff04a29c2137df5/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..5f02688ef4bdaac474b3433117f9ad3bb267d1b7
--- /dev/null
+++ b/records/sd1/sweep/54875ff04a29c2137df5/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "54875ff04a29c2137df5",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:01Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.3.0.op",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "in.hi",
+   "kind": "downsample",
+   "module_type": "Conv2d",
+   "container": "Downsample",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/64a8d5fb0289ef44985bc7848b41aeae507d3bfc5789c7d25886add4fa427f4a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "64a8d5fb0289ef44985bc7848b41aeae507d3bfc5789c7d25886add4fa427f4a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.Downsample.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "76e870a45c7fefb178cc"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54875ff04a29c2137df5/workflow.json b/records/sd1/sweep/54875ff04a29c2137df5/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..18550dbe785675a72cce51ac820060b33a5b2e77
--- /dev/null
+++ b/records/sd1/sweep/54875ff04a29c2137df5/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54875ff04a29c2137df5)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.3.0.op\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/549ed83bc2c78cb34090/measurements.json b/records/sd1/sweep/549ed83bc2c78cb34090/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..91c1c40986b3219eaf70b824f7377deeadf5a31d
--- /dev/null
+++ b/records/sd1/sweep/549ed83bc2c78cb34090/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "549ed83bc2c78cb34090",
+ "values": {
+  "cosine_distance": {
+   "value": 0.361887,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.490451,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.171613,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.116163,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 39.916,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00125,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 41.335,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0322,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3648,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/549ed83bc2c78cb34090/record.json b/records/sd1/sweep/549ed83bc2c78cb34090/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d20e322494adc5b79f25f81df417c4b1703350f9
--- /dev/null
+++ b/records/sd1/sweep/549ed83bc2c78cb34090/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "549ed83bc2c78cb34090",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:39:40Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.1.0.out_layers.3",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "in.hi",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/40604033603468ba46bb3115d9567b5fdd889628a74a87c6b3de0d3fbecd243a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "40604033603468ba46bb3115d9567b5fdd889628a74a87c6b3de0d3fbecd243a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "f02556fc891265507d25"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/549ed83bc2c78cb34090/workflow.json b/records/sd1/sweep/549ed83bc2c78cb34090/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..b7c72f6708f300fc143931cb924ace4e4cfecdb6
--- /dev/null
+++ b/records/sd1/sweep/549ed83bc2c78cb34090/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 549ed83bc2c78cb34090)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.1.0.out_layers.3\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54a4914e1b7f6892b7cb/interpretations.jsonl b/records/sd1/sweep/54a4914e1b7f6892b7cb/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..17590eb54603e7d546bd506d39f72b9d1ce5ab32
--- /dev/null
+++ b/records/sd1/sweep/54a4914e1b7f6892b7cb/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "54a4914e1b7f6892b7cb"}, "kind": "caption", "value": "Smeared, painterly blur of blue, red and orange suggesting a neon street", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:51:21Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a176acfb10018cc1"}
+{"target": {"record": "54a4914e1b7f6892b7cb"}, "kind": "change", "value": "The street melts into a soft, brushy smear; the blue centre and warm sides survive as colour fields while the buildings and people dissolve.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:51:21Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "5b9225a5d7a81773"}
+{"target": {"record": "54a4914e1b7f6892b7cb"}, "kind": "keywords", "value": ["smeared neon street", "painterly blur", "blue and orange", "dissolved city"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:51:21Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "7af20f3d3d0c1ef1"}
+{"target": {"record": "54a4914e1b7f6892b7cb"}, "kind": "effect_tags", "value": ["abstract", "blur", "painterly", "fragmented", "subject_lost"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T04:51:21Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "0aa50c17ad658e6d"}
diff --git a/records/sd1/sweep/54a4914e1b7f6892b7cb/measurements.json b/records/sd1/sweep/54a4914e1b7f6892b7cb/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..7cfb66001782a18f7e0a5ccc620494cd99426ce9
--- /dev/null
+++ b/records/sd1/sweep/54a4914e1b7f6892b7cb/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "54a4914e1b7f6892b7cb",
+ "values": {
+  "std": {
+   "value": 28.912,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0077,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 23.755,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.14727,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.523677,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.974442,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.318175,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.6556,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.6799,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54a4914e1b7f6892b7cb/record.json b/records/sd1/sweep/54a4914e1b7f6892b7cb/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..d6799e59ecf0b5c456079fbff6a64fee1a3832f0
--- /dev/null
+++ b/records/sd1/sweep/54a4914e1b7f6892b7cb/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "54a4914e1b7f6892b7cb",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:10:17Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30.0
+   },
+   "group": "out.hi",
+   "kind": "attn_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "slight",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.hi",
+   "rotate",
+   30.0,
+   "late"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 30.0,
+    "applied": 30.0,
+    "reason": "outside safe range [0.0, 5.0] for out.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e7a3b27dfe7e4b6eb83d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54a4914e1b7f6892b7cb/workflow.json b/records/sd1/sweep/54a4914e1b7f6892b7cb/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..b731ecb889ddeb181a796cfb864838a7c151f35a
--- /dev/null
+++ b/records/sd1/sweep/54a4914e1b7f6892b7cb/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54a4914e1b7f6892b7cb)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30.0\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54be575ca13f7f088fa0/measurements.json b/records/sd1/sweep/54be575ca13f7f088fa0/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e05b7e1e5a9181bc91b9c750f0b70c7ee0f94f93
--- /dev/null
+++ b/records/sd1/sweep/54be575ca13f7f088fa0/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "54be575ca13f7f088fa0",
+ "values": {
+  "cosine_distance": {
+   "value": 0.114793,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.137987,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.089327,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.120707,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 63.806,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00393,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 14.388,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9494,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5073,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54be575ca13f7f088fa0/record.json b/records/sd1/sweep/54be575ca13f7f088fa0/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d7f70d1096689f0b94395f51c758ab8708dd5a0
--- /dev/null
+++ b/records/sd1/sweep/54be575ca13f7f088fa0/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "54be575ca13f7f088fa0",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:16Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.1.proj_out",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.mid",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/510357f1d04678398ac7ec4c7759443ff4c04508e5bbeaa3c06bea8f8fb27dc1.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "510357f1d04678398ac7ec4c7759443ff4c04508e5bbeaa3c06bea8f8fb27dc1",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "1a872c7307beca843246"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54be575ca13f7f088fa0/workflow.json b/records/sd1/sweep/54be575ca13f7f088fa0/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..1b5d42ce956750a31d5b4f3d29a76306d8574619
--- /dev/null
+++ b/records/sd1/sweep/54be575ca13f7f088fa0/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54be575ca13f7f088fa0)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.1.proj_out\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54d1894068092bee6d62/measurements.json b/records/sd1/sweep/54d1894068092bee6d62/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..aee0ea0423764efc28f531ed0b818315d58df59e
--- /dev/null
+++ b/records/sd1/sweep/54d1894068092bee6d62/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "54d1894068092bee6d62",
+ "values": {
+  "cosine_distance": {
+   "value": 0.219572,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.480615,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.421708,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.231317,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 45.768,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00503,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 26.801,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.1126,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5349,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54d1894068092bee6d62/record.json b/records/sd1/sweep/54d1894068092bee6d62/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1ab5f73cf9a5b0816f348a27796c613987693c62
--- /dev/null
+++ b/records/sd1/sweep/54d1894068092bee6d62/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "54d1894068092bee6d62",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:50:12Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.in_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/1579b37d92c30c0626e22fa5f58603e80047c19a56a5aa6e548cb4f9349ff3a0.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "1579b37d92c30c0626e22fa5f58603e80047c19a56a5aa6e548cb4f9349ff3a0",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a03cf722ebff2f535aef"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54d1894068092bee6d62/workflow.json b/records/sd1/sweep/54d1894068092bee6d62/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..5d4ea609a1df5beae5741de1bfedda385313e0e9
--- /dev/null
+++ b/records/sd1/sweep/54d1894068092bee6d62/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54d1894068092bee6d62)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.in_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54d2cce125d9aff87680/measurements.json b/records/sd1/sweep/54d2cce125d9aff87680/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..824c5eaaa63f25140c73ff90233901f17c2b7318
--- /dev/null
+++ b/records/sd1/sweep/54d2cce125d9aff87680/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "54d2cce125d9aff87680",
+ "values": {
+  "cosine_distance": {
+   "value": 0.157688,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.172806,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.094666,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.072316,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 65.191,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00351,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 17.441,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0081,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1962,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54d2cce125d9aff87680/record.json b/records/sd1/sweep/54d2cce125d9aff87680/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b6434e9530eb123be04e2dff433e829a5aeeb237
--- /dev/null
+++ b/records/sd1/sweep/54d2cce125d9aff87680/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "54d2cce125d9aff87680",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:02Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.1.0.skip_connection",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.5
+   },
+   "group": "out.lo",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/12c49cc5a77e5898e2431ea17d4fba17ef34b42b50e5bae615d8efff077ab325.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "12c49cc5a77e5898e2431ea17d4fba17ef34b42b50e5bae615d8efff077ab325",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "9ac1d09601d454db9388"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54d2cce125d9aff87680/workflow.json b/records/sd1/sweep/54d2cce125d9aff87680/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..b32b0f13530660f5c55a177ae6c41faf362f6bf9
--- /dev/null
+++ b/records/sd1/sweep/54d2cce125d9aff87680/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54d2cce125d9aff87680)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.1.0.skip_connection\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54e25ddd8a09e6f1a64e/interpretations.jsonl b/records/sd1/sweep/54e25ddd8a09e6f1a64e/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..b05b3d09b0ef822eb37a7e377a86eaa649d41ca5
--- /dev/null
+++ b/records/sd1/sweep/54e25ddd8a09e6f1a64e/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "54e25ddd8a09e6f1a64e"}, "kind": "caption", "value": "Studio portrait of bob-haired woman in black top against grey card, practically identical", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:17:21Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "7696988844009354"}
+{"target": {"record": "54e25ddd8a09e6f1a64e"}, "kind": "change", "value": "Almost no visible change apart from slightly warmer lips and skin.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:17:21Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "d07de295c16bc8ff"}
+{"target": {"record": "54e25ddd8a09e6f1a64e"}, "kind": "keywords", "value": ["near identical", "studio portrait", "minimal change", "warm lips"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:17:21Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "84bb35fa666c8129"}
+{"target": {"record": "54e25ddd8a09e6f1a64e"}, "kind": "effect_tags", "value": ["subtle", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:17:21Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "04881a90fa3e0310"}
diff --git a/records/sd1/sweep/54e25ddd8a09e6f1a64e/measurements.json b/records/sd1/sweep/54e25ddd8a09e6f1a64e/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..fc49d0081adef016ab9bffd46334205acc7a6cb1
--- /dev/null
+++ b/records/sd1/sweep/54e25ddd8a09e6f1a64e/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "54e25ddd8a09e6f1a64e",
+ "values": {
+  "cosine_distance": {
+   "value": 0.0,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.0,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": -0.0,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 66.477,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00637,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.0,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0093,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2957,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54e25ddd8a09e6f1a64e/record.json b/records/sd1/sweep/54e25ddd8a09e6f1a64e/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..60a354070231c7fd0849d3e1b6fb5659cedc8700
--- /dev/null
+++ b/records/sd1/sweep/54e25ddd8a09e6f1a64e/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "54e25ddd8a09e6f1a64e",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:52:24Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.0.h_upd",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.5
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "shrink"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/5dd9b8fb42253d2a6d3fbc7f61f2be364228dd1137e06bac465f8578fc85c5f2.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "5dd9b8fb42253d2a6d3fbc7f61f2be364228dd1137e06bac465f8578fc85c5f2",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "6134d35db3e8c8b2c0b0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54e25ddd8a09e6f1a64e/workflow.json b/records/sd1/sweep/54e25ddd8a09e6f1a64e/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..bd5721afebc4a5e9d6b35d42a266f88d564b6d07
--- /dev/null
+++ b/records/sd1/sweep/54e25ddd8a09e6f1a64e/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54e25ddd8a09e6f1a64e)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.0.h_upd\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54e330c1cd0cc04b428a/measurements.json b/records/sd1/sweep/54e330c1cd0cc04b428a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..da38144f5294bbd35889d27f554cd860ea31575e
--- /dev/null
+++ b/records/sd1/sweep/54e330c1cd0cc04b428a/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "54e330c1cd0cc04b428a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.375856,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.403879,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.170631,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.120263,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 50.614,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00554,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 38.43,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9961,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3855,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54e330c1cd0cc04b428a/record.json b/records/sd1/sweep/54e330c1cd0cc04b428a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..1e749503a412045a8b21d96de8fc92d2305791dd
--- /dev/null
+++ b/records/sd1/sweep/54e330c1cd0cc04b428a/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "54e330c1cd0cc04b428a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:56Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "middle_block.1.proj_out",
+   "op": "scale",
+   "args": {
+    "scale_factor": 2
+   },
+   "group": "mid",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "strong_grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/db40805d5f39093eddc66b09065d9490876e47f28f337220ab9ee13f6667fba9.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "db40805d5f39093eddc66b09065d9490876e47f28f337220ab9ee13f6667fba9",
+  "type_path": "UNetModel.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "9f1e527dffb2eac6bfe0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54e330c1cd0cc04b428a/workflow.json b/records/sd1/sweep/54e330c1cd0cc04b428a/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..73941d0d570e26bcf2c82fb4c6f690cae0074249
--- /dev/null
+++ b/records/sd1/sweep/54e330c1cd0cc04b428a/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54e330c1cd0cc04b428a)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1.proj_out\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54eb10d89bf65954b1bb/interpretations.jsonl b/records/sd1/sweep/54eb10d89bf65954b1bb/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..30bee652f60a6af617afdd3c1b75a2297f4126e3
--- /dev/null
+++ b/records/sd1/sweep/54eb10d89bf65954b1bb/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "54eb10d89bf65954b1bb"}, "kind": "caption", "value": "Dense chaotic mosaic of saturated cyan, red, white and black blotches with no subject", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:17:40Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ec2594c09b4dc704"}
+{"target": {"record": "54eb10d89bf65954b1bb"}, "kind": "change", "value": "The portrait is replaced by a vivid, glitchy confetti of saturated colour fragments over black, with no recognisable form.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:17:40Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "e02efca6cfb86ace"}
+{"target": {"record": "54eb10d89bf65954b1bb"}, "kind": "keywords", "value": ["colour confetti", "glitch noise", "saturated fragments", "abstract mosaic", "lost subject"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:17:40Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "6205ea72c1598a36"}
+{"target": {"record": "54eb10d89bf65954b1bb"}, "kind": "effect_tags", "value": ["abstract", "subject_lost", "saturated", "fragmented", "glitch", "degenerate", "high_contrast"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T08:17:40Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "94e5d6dfe9cb3a61"}
diff --git a/records/sd1/sweep/54eb10d89bf65954b1bb/measurements.json b/records/sd1/sweep/54eb10d89bf65954b1bb/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..0f753ec2586ff4e9a7a8ee820f5251e585aed1df
--- /dev/null
+++ b/records/sd1/sweep/54eb10d89bf65954b1bb/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "54eb10d89bf65954b1bb",
+ "values": {
+  "cosine_distance": {
+   "value": 0.66589,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.74954,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.929331,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.499984,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 92.355,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01221,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 87.925,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8701,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.8587,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54eb10d89bf65954b1bb/record.json b/records/sd1/sweep/54eb10d89bf65954b1bb/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..29bd1ee88ea1d817e1466a3195201fec2776393b
--- /dev/null
+++ b/records/sd1/sweep/54eb10d89bf65954b1bb/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "54eb10d89bf65954b1bb",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:23Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.0.skip_connection",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.hi",
+   "kind": "skip",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/5238ba73ca540cfdf842006cca1082dd122181cc063ea769b7583165cb72fc0b.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "5238ba73ca540cfdf842006cca1082dd122181cc063ea769b7583165cb72fc0b",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "7366abd5679f07b1e682"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54eb10d89bf65954b1bb/workflow.json b/records/sd1/sweep/54eb10d89bf65954b1bb/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..8d99e5908b29420399d3d1ba19ffc9035c3f701f
--- /dev/null
+++ b/records/sd1/sweep/54eb10d89bf65954b1bb/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54eb10d89bf65954b1bb)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.0.skip_connection\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54f00881dfdb208e4fe6/measurements.json b/records/sd1/sweep/54f00881dfdb208e4fe6/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..1b93e5168cf120eade8fb6e8a5b44607b9daee32
--- /dev/null
+++ b/records/sd1/sweep/54f00881dfdb208e4fe6/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "54f00881dfdb208e4fe6",
+ "values": {
+  "cosine_distance": {
+   "value": 0.277885,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.39187,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.199158,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.099694,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 37.645,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0025,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 33.776,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9543,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0569,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54f00881dfdb208e4fe6/record.json b/records/sd1/sweep/54f00881dfdb208e4fe6/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..85c6e085f0a4dcd9a523bc5142670398a1acd493
--- /dev/null
+++ b/records/sd1/sweep/54f00881dfdb208e4fe6/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "54f00881dfdb208e4fe6",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:47:20Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.11.0.emb_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.hi",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/91aa22e8eeec634b2eed8205422a43f09dfe783f6d5fee48c20a362db141a50d.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "91aa22e8eeec634b2eed8205422a43f09dfe783f6d5fee48c20a362db141a50d",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "02eae618616251fb87c0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54f00881dfdb208e4fe6/workflow.json b/records/sd1/sweep/54f00881dfdb208e4fe6/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f53d3790ad58dfff430e3750d552af2db58ec9e7
--- /dev/null
+++ b/records/sd1/sweep/54f00881dfdb208e4fe6/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54f00881dfdb208e4fe6)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.11.0.emb_layers.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54f55c13cb0821c8217d/measurements.json b/records/sd1/sweep/54f55c13cb0821c8217d/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c3405d32fddb5a6fb050604413f45b76acee1862
--- /dev/null
+++ b/records/sd1/sweep/54f55c13cb0821c8217d/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "54f55c13cb0821c8217d",
+ "values": {
+  "std": {
+   "value": 32.265,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01953,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.774,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.000128,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.000737,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.000584,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.00022,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0045,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2249,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54f55c13cb0821c8217d/record.json b/records/sd1/sweep/54f55c13cb0821c8217d/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3d0a19f8180cfd17f7ee893c314495c949ef563e
--- /dev/null
+++ b/records/sd1/sweep/54f55c13cb0821c8217d/record.json
@@ -0,0 +1,79 @@
+{
+ "id": "54f55c13cb0821c8217d",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:23:31Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "add_noise",
+   "args": {
+    "noise_std": 0.8,
+    "seed": 1234
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "heavy",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "add_noise",
+   0.8,
+   "late"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "275b7f34827ecf78dc91"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/54f55c13cb0821c8217d/workflow.json b/records/sd1/sweep/54f55c13cb0821c8217d/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c607fa2007ddedbe2da048ef15fdcaddc0fa8aa8
--- /dev/null
+++ b/records/sd1/sweep/54f55c13cb0821c8217d/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 54f55c13cb0821c8217d)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"add_noise\",\n   \"module_args\": {\n    \"noise_std\": 0.8,\n    \"seed\": 1234\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550ae85cbb7dbb7e8bf1/measurements.json b/records/sd1/sweep/550ae85cbb7dbb7e8bf1/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6a1911f3ea9426f76ca1e98d554ebdc0d55383c5
--- /dev/null
+++ b/records/sd1/sweep/550ae85cbb7dbb7e8bf1/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "550ae85cbb7dbb7e8bf1",
+ "values": {
+  "std": {
+   "value": 39.323,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.01898,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 25.36,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.245288,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.390339,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.153841,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.056815,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9783,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0417,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550ae85cbb7dbb7e8bf1/record.json b/records/sd1/sweep/550ae85cbb7dbb7e8bf1/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..52c544ba37272352bc4e3af1b53f93f2eec13232
--- /dev/null
+++ b/records/sd1/sweep/550ae85cbb7dbb7e8bf1/record.json
@@ -0,0 +1,76 @@
+{
+ "id": "550ae85cbb7dbb7e8bf1",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:07:49Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "scale",
+   1.4,
+   "all"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "62e397272b07f6669ba3"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550ae85cbb7dbb7e8bf1/workflow.json b/records/sd1/sweep/550ae85cbb7dbb7e8bf1/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..1c505a3cfacda6edc1b9ec0987ed93a4b337f7fa
--- /dev/null
+++ b/records/sd1/sweep/550ae85cbb7dbb7e8bf1/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 550ae85cbb7dbb7e8bf1)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550c5ace85226c555637/interpretations.jsonl b/records/sd1/sweep/550c5ace85226c555637/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..bf183f056461fb4ef9f496e1a67ae54e2f67e5b3
--- /dev/null
+++ b/records/sd1/sweep/550c5ace85226c555637/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "550c5ace85226c555637"}, "kind": "caption", "value": "Woman with dark wavy bob in front of a blocky, patchy grey-green background", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:20:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "e65a344022309260"}
+{"target": {"record": "550c5ace85226c555637"}, "kind": "change", "value": "The background breaks into a patchwork of small rectangular grey and green blocks while the face softens and the hair turns darker and wavier.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:20:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "bd2e5ccf6a495778"}
+{"target": {"record": "550c5ace85226c555637"}, "kind": "keywords", "value": ["patchwork background", "blocky texture", "dark wavy bob", "soft face"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:20:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "22818a45c355acd3"}
+{"target": {"record": "550c5ace85226c555637"}, "kind": "effect_tags", "value": ["subject_kept", "texture", "pattern", "painterly"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:20:58Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "42ef461c63c7f160"}
diff --git a/records/sd1/sweep/550c5ace85226c555637/measurements.json b/records/sd1/sweep/550c5ace85226c555637/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..cb3f4681f890a580ab4ac95cfe4e87fd22bd9a25
--- /dev/null
+++ b/records/sd1/sweep/550c5ace85226c555637/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "550c5ace85226c555637",
+ "values": {
+  "cosine_distance": {
+   "value": 0.31098,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.360355,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.213489,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.081713,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 56.338,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00442,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 35.774,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9675,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0962,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550c5ace85226c555637/record.json b/records/sd1/sweep/550c5ace85226c555637/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..4eeb4266c6510198a188aee2288d166570fe6ae7
--- /dev/null
+++ b/records/sd1/sweep/550c5ace85226c555637/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "550c5ace85226c555637",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:48:08Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.4.0.out_layers.3",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "in.mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/a7691e96594ae7d47bac8ecc554594d4ebd6064aa8819b5c5f02f3ca037b9644.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "a7691e96594ae7d47bac8ecc554594d4ebd6064aa8819b5c5f02f3ca037b9644",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "b9a0f21535d0c8690860"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550c5ace85226c555637/workflow.json b/records/sd1/sweep/550c5ace85226c555637/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e7127310fedcc6d62b8b5ae6ad71d55a8204b6e
--- /dev/null
+++ b/records/sd1/sweep/550c5ace85226c555637/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 550c5ace85226c555637)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.4.0.out_layers.3\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550d73c67b9d32d11727/measurements.json b/records/sd1/sweep/550d73c67b9d32d11727/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6f8c101c71cb45132869a5ebe165d36bcca151ed
--- /dev/null
+++ b/records/sd1/sweep/550d73c67b9d32d11727/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "550d73c67b9d32d11727",
+ "values": {
+  "std": {
+   "value": 49.361,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00436,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": true,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": [
+    "clip_degenerate"
+   ]
+  },
+  "mae_vs_baseline": {
+   "value": 49.52,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.466769,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.746123,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 1.01479,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.365701,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.5972,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.9607,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550d73c67b9d32d11727/record.json b/records/sd1/sweep/550d73c67b9d32d11727/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..6bfa3702ab43c0fd5c04424e42e998ec91a863c6
--- /dev/null
+++ b/records/sd1/sweep/550d73c67b9d32d11727/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "550d73c67b9d32d11727",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:14:16Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 1.4
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "grow"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "scale",
+   1.4,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 1.4,
+    "applied": 1.4,
+    "reason": "outside safe range [0.95, 1.05] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e225637ea28cf216187f"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/550d73c67b9d32d11727/workflow.json b/records/sd1/sweep/550d73c67b9d32d11727/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..692fc6836c9cc75c536cf8963929dd3ad1bb7ba1
--- /dev/null
+++ b/records/sd1/sweep/550d73c67b9d32d11727/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 550d73c67b9d32d11727)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 1.4\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55219d1ca957d8db7d64/measurements.json b/records/sd1/sweep/55219d1ca957d8db7d64/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..fa63c14ac90477373ae63562188b937127ed5a3a
--- /dev/null
+++ b/records/sd1/sweep/55219d1ca957d8db7d64/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "55219d1ca957d8db7d64",
+ "values": {
+  "std": {
+   "value": 32.277,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0276,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 19.889,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.135129,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.251245,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.075359,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.022786,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.058,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3093,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55219d1ca957d8db7d64/record.json b/records/sd1/sweep/55219d1ca957d8db7d64/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..ae1775443c8b434ca2fa5855b1ae543126ef283e
--- /dev/null
+++ b/records/sd1/sweep/55219d1ca957d8db7d64/record.json
@@ -0,0 +1,78 @@
+{
+ "id": "55219d1ca957d8db7d64",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:17:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "add_scalar",
+   "args": {
+    "scalar": 0.8
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 0.333,
+   "window": "early",
+   "bucket": "mid",
+   "steps_min": 0,
+   "steps_max": 1
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "add_scalar",
+   0.8,
+   "early"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "4ac83071e6bb9985a9c6"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55219d1ca957d8db7d64/workflow.json b/records/sd1/sweep/55219d1ca957d8db7d64/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..b986bac5af26abd54f3ac23d7c10d82a404435fe
--- /dev/null
+++ b/records/sd1/sweep/55219d1ca957d8db7d64/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55219d1ca957d8db7d64)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"add_scalar\",\n   \"module_args\": {\n    \"scalar\": 0.8\n   }\n  }\n ],\n \"steps_min\": 0,\n \"steps_max\": 1\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55261a0ab014d20808d0/measurements.json b/records/sd1/sweep/55261a0ab014d20808d0/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c842a8a976a3595f716276637ce9a1620fa55267
--- /dev/null
+++ b/records/sd1/sweep/55261a0ab014d20808d0/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "55261a0ab014d20808d0",
+ "values": {
+  "std": {
+   "value": 27.224,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0176,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 28.76,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.384817,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.586438,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.610233,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.218122,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.885,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0169,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55261a0ab014d20808d0/record.json b/records/sd1/sweep/55261a0ab014d20808d0/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..9a62a2fabfd974ed1db715d5666d517a328613e9
--- /dev/null
+++ b/records/sd1/sweep/55261a0ab014d20808d0/record.json
@@ -0,0 +1,77 @@
+{
+ "id": "55261a0ab014d20808d0",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-02T02:11:48Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a lighthouse on a rocky coast at dusk, oil painting",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e368f07d8e06f0e7"
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0",
+   "op": "scale",
+   "args": {
+    "scale_factor": 0.7
+   },
+   "group": "out.mid",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "shrink",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "de1eccace874429a61dc",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "out.mid",
+   "scale",
+   0.7,
+   "middle"
+  ],
+  "clamps": [],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ]
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55261a0ab014d20808d0/workflow.json b/records/sd1/sweep/55261a0ab014d20808d0/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a81f289247caaf53efa378a892f3e0b3bfd9ec41
--- /dev/null
+++ b/records/sd1/sweep/55261a0ab014d20808d0/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55261a0ab014d20808d0)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0\",\n   \"module_type\": \"scale\",\n   \"module_args\": {\n    \"scale_factor\": 0.7\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a lighthouse on a rocky coast at dusk, oil painting"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/552c45d75640f81d6ff9/measurements.json b/records/sd1/sweep/552c45d75640f81d6ff9/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..1561914dc1b495e8f3778ebc9bdc671402e6048a
--- /dev/null
+++ b/records/sd1/sweep/552c45d75640f81d6ff9/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "552c45d75640f81d6ff9",
+ "values": {
+  "std": {
+   "value": 24.071,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0118,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 41.688,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.374581,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.724191,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.917611,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.341334,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.6969,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.6942,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/552c45d75640f81d6ff9/record.json b/records/sd1/sweep/552c45d75640f81d6ff9/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3675a2facf06ee6c39a2cb903bbdd52ab0ba8
--- /dev/null
+++ b/records/sd1/sweep/552c45d75640f81d6ff9/record.json
@@ -0,0 +1,83 @@
+{
+ "id": "552c45d75640f81d6ff9",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T22:52:38Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.0",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "in.mid",
+   "kind": "res_block",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "1bbda6ca1f7efc631dfb",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.mid",
+   "multiply",
+   0.0,
+   "all"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for in.mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "a8a5a2e4335d9b611574"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/552c45d75640f81d6ff9/workflow.json b/records/sd1/sweep/552c45d75640f81d6ff9/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..b80ac54ae16cc8d1e43bbbff37381dc2d0119ea6
--- /dev/null
+++ b/records/sd1/sweep/552c45d75640f81d6ff9/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 552c45d75640f81d6ff9)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5532cc0ccf1951b10626/measurements.json b/records/sd1/sweep/5532cc0ccf1951b10626/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e0df4d5bf08aeb7724ae0bbb4a9ea6045ad7533d
--- /dev/null
+++ b/records/sd1/sweep/5532cc0ccf1951b10626/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5532cc0ccf1951b10626",
+ "values": {
+  "cosine_distance": {
+   "value": 0.552597,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.588879,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.622161,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.228383,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 35.168,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00235,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 49.199,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0663,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1733,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5532cc0ccf1951b10626/record.json b/records/sd1/sweep/5532cc0ccf1951b10626/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..90bbbef00c387828a5bf0120e3073068b99ca22f
--- /dev/null
+++ b/records/sd1/sweep/5532cc0ccf1951b10626/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5532cc0ccf1951b10626",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:49:39Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.0.out_layers.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "GroupNorm",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/a8f082ab407f2a74ecee8c4929554ae6f467bada9aac2fd991bd6430d4b915ec.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "a8f082ab407f2a74ecee8c4929554ae6f467bada9aac2fd991bd6430d4b915ec",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "2e0fec1f1af79c5a33b1"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5532cc0ccf1951b10626/workflow.json b/records/sd1/sweep/5532cc0ccf1951b10626/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..864c8e75ff8b01dc9387acf5f48585c83ebc45e2
--- /dev/null
+++ b/records/sd1/sweep/5532cc0ccf1951b10626/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5532cc0ccf1951b10626)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.0.out_layers.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55374a67a0fd2f2631fd/measurements.json b/records/sd1/sweep/55374a67a0fd2f2631fd/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..15dc4abf4c7664ad1819aea9b0037a0eadaf7d85
--- /dev/null
+++ b/records/sd1/sweep/55374a67a0fd2f2631fd/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "55374a67a0fd2f2631fd",
+ "values": {
+  "cosine_distance": {
+   "value": 0.542962,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.709987,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.610642,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.300121,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 21.514,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00989,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 47.574,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0468,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.5048,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55374a67a0fd2f2631fd/record.json b/records/sd1/sweep/55374a67a0fd2f2631fd/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3bfb09b359f2c64c9d5334c764a494f7c8e7e92e
--- /dev/null
+++ b/records/sd1/sweep/55374a67a0fd2f2631fd/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "55374a67a0fd2f2631fd",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:49:16Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.0.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 90
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/8825ac1a626d4fb0cb2b37a6698d2ffcb2a2b32be16483873809230758c6df45.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "8825ac1a626d4fb0cb2b37a6698d2ffcb2a2b32be16483873809230758c6df45",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "02ad6611733b3c06ce91"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55374a67a0fd2f2631fd/workflow.json b/records/sd1/sweep/55374a67a0fd2f2631fd/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..a7aa9f87a2409a6b48e15170d4d70e48b2842bb9
--- /dev/null
+++ b/records/sd1/sweep/55374a67a0fd2f2631fd/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55374a67a0fd2f2631fd)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.0.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 90\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/555a1082f29f0ddfe708/measurements.json b/records/sd1/sweep/555a1082f29f0ddfe708/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..20d23358ec78131d910e079df55220893a577c35
--- /dev/null
+++ b/records/sd1/sweep/555a1082f29f0ddfe708/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "555a1082f29f0ddfe708",
+ "values": {
+  "std": {
+   "value": 32.217,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.02096,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 9.158,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.031389,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.086029,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.073229,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.011917,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 1.0077,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2244,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/555a1082f29f0ddfe708/record.json b/records/sd1/sweep/555a1082f29f0ddfe708/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7eeb886eede442bff052e5555ce24c9300ec58a1
--- /dev/null
+++ b/records/sd1/sweep/555a1082f29f0ddfe708/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "555a1082f29f0ddfe708",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T23:20:56Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 1234,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a city street at night with neon signs, digital illustration",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "e39807168df36de8"
+ },
+ "bends": [
+  {
+   "path": "middle_block.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.0
+   },
+   "group": "mid",
+   "kind": "attn_block",
+   "start": 0.667,
+   "end": 1.0,
+   "window": "late",
+   "bucket": "zero",
+   "steps_min": 4,
+   "steps_max": 5
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "81ef55543b3b3c080780",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "mid",
+   "multiply",
+   0.0,
+   "late"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 0.0,
+    "applied": 0.0,
+    "reason": "outside safe range [0.5, 2.0] for mid (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "3bcf6969b4426802b26a"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/555a1082f29f0ddfe708/workflow.json b/records/sd1/sweep/555a1082f29f0ddfe708/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..48e033ecf3f6ddc0249276c5981292d6cdcaa366
--- /dev/null
+++ b/records/sd1/sweep/555a1082f29f0ddfe708/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 555a1082f29f0ddfe708)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"middle_block.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.0\n   }\n  }\n ],\n \"steps_min\": 4,\n \"steps_max\": 5\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a city street at night with neon signs, digital illustration"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 1234,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55665a2ac46340211f74/measurements.json b/records/sd1/sweep/55665a2ac46340211f74/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6fb23c761bb9cae5be1c6c4ef0000d53a1480870
--- /dev/null
+++ b/records/sd1/sweep/55665a2ac46340211f74/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "55665a2ac46340211f74",
+ "values": {
+  "cosine_distance": {
+   "value": 0.74572,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.669318,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.744031,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.423584,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 27.106,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00608,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 65.123,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8288,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.2588,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55665a2ac46340211f74/record.json b/records/sd1/sweep/55665a2ac46340211f74/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..0c0e9e4f95dc496b8a83cec0d1aabaa34c228394
--- /dev/null
+++ b/records/sd1/sweep/55665a2ac46340211f74/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "55665a2ac46340211f74",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:15Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.0.out_layers.2",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "out.lo",
+   "kind": "res_branch",
+   "module_type": "Dropout",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/3b789c987087cfd9b7b131374908179060e54a8d1dd60af10fd74443f6069057.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "3b789c987087cfd9b7b131374908179060e54a8d1dd60af10fd74443f6069057",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "30e4c888165b5f62df51"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55665a2ac46340211f74/workflow.json b/records/sd1/sweep/55665a2ac46340211f74/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..c99ac9890e25df5690c665e9eb85c175de0c0b7b
--- /dev/null
+++ b/records/sd1/sweep/55665a2ac46340211f74/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55665a2ac46340211f74)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.0.out_layers.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5574d44994086ee8dfed/measurements.json b/records/sd1/sweep/5574d44994086ee8dfed/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..2a041f8fef3b1d3cbde33f08d8f619f74ef09534
--- /dev/null
+++ b/records/sd1/sweep/5574d44994086ee8dfed/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5574d44994086ee8dfed",
+ "values": {
+  "cosine_distance": {
+   "value": 0.098688,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.145305,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.041659,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.029372,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 52.993,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00599,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 15.756,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9842,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0755,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5574d44994086ee8dfed/record.json b/records/sd1/sweep/5574d44994086ee8dfed/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b5a6de386c734066f7851f101b919904bcec0b2d
--- /dev/null
+++ b/records/sd1/sweep/5574d44994086ee8dfed/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5574d44994086ee8dfed",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:53Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.in_layers.2",
+   "op": "multiply",
+   "args": {
+    "scalar": 0.5
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "Conv2d",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "damp"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/592a64f23df39378cc5f5c9ee095a8ba2eb69bc05d42e3f25034a78131176e8b.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "592a64f23df39378cc5f5c9ee095a8ba2eb69bc05d42e3f25034a78131176e8b",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "3cd0424a5c728b43d519"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5574d44994086ee8dfed/workflow.json b/records/sd1/sweep/5574d44994086ee8dfed/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..8451dceb490097e961d3ae0587f57c1350653c57
--- /dev/null
+++ b/records/sd1/sweep/5574d44994086ee8dfed/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5574d44994086ee8dfed)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.in_layers.2\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5584a8464156d33cf881/interpretations.jsonl b/records/sd1/sweep/5584a8464156d33cf881/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..4f8560acd6ae28ba50810de0a6908e0c0ab50326
--- /dev/null
+++ b/records/sd1/sweep/5584a8464156d33cf881/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "5584a8464156d33cf881"}, "kind": "caption", "value": "Illustrated portrait of a dark-haired woman in black turtleneck, slightly cooler and smoother", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:33:49Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "6f9c4351d7481978"}
+{"target": {"record": "5584a8464156d33cf881"}, "kind": "change", "value": "Almost no visible change, apart from a slightly smoother face and a marginally cooler grey background.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:33:49Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "db5ab7558c60e0a7"}
+{"target": {"record": "5584a8464156d33cf881"}, "kind": "keywords", "value": ["subtle variation", "same portrait", "cool grey"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:33:49Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "8c313a700eb32cd9"}
+{"target": {"record": "5584a8464156d33cf881"}, "kind": "effect_tags", "value": ["subject_kept", "subtle"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:33:49Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "0ceb0b23269f09f1"}
diff --git a/records/sd1/sweep/5584a8464156d33cf881/measurements.json b/records/sd1/sweep/5584a8464156d33cf881/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..319a098b3276a67bb7b37be513cd7676fbc8bcdf
--- /dev/null
+++ b/records/sd1/sweep/5584a8464156d33cf881/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "5584a8464156d33cf881",
+ "values": {
+  "cosine_distance": {
+   "value": 0.0,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.0,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.0,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 60.618,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00534,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 0.0,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1553,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5584a8464156d33cf881/record.json b/records/sd1/sweep/5584a8464156d33cf881/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..23a5f1b36be60f13eee2c73981ed8f23f9bb08ec
--- /dev/null
+++ b/records/sd1/sweep/5584a8464156d33cf881/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "5584a8464156d33cf881",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:50:05Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.6.0.h_upd",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 180
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "Identity",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "half"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/fb7a14b8973e132f7a3e821b0d04a54e3aa353bc52207de5d5113db7495d2ec6.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "fb7a14b8973e132f7a3e821b0d04a54e3aa353bc52207de5d5113db7495d2ec6",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Identity",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "3f40f506eb65baecb00b"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5584a8464156d33cf881/workflow.json b/records/sd1/sweep/5584a8464156d33cf881/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..e6cf9dd18b6c2d2f0fd67528d0cfccc8d536f7d7
--- /dev/null
+++ b/records/sd1/sweep/5584a8464156d33cf881/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5584a8464156d33cf881)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.6.0.h_upd\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 180\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5585bce436370c177b2f/measurements.json b/records/sd1/sweep/5585bce436370c177b2f/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..667b4308779b583f13b1eaa8a44b6a133f14bb1f
--- /dev/null
+++ b/records/sd1/sweep/5585bce436370c177b2f/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "5585bce436370c177b2f",
+ "values": {
+  "std": {
+   "value": 44.63,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00342,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 45.676,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "cosine_distance": {
+   "value": 0.543007,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.683542,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.769218,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.356683,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "prompt_retention": {
+   "value": 0.9085,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4504,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5585bce436370c177b2f/record.json b/records/sd1/sweep/5585bce436370c177b2f/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3b63db4f5343deff5085968291da665e7f1640b6
--- /dev/null
+++ b/records/sd1/sweep/5585bce436370c177b2f/record.json
@@ -0,0 +1,85 @@
+{
+ "id": "5585bce436370c177b2f",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-04T21:58:57Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 7,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "a still life of fruit and a glass vase on a wooden table, watercolor",
+  "negative": "lowres, watermark, text",
+  "prompt_key": "2360b104bf93fdae"
+ },
+ "bends": [
+  {
+   "path": "input_blocks.2.0",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30.0
+   },
+   "group": "in.hi",
+   "kind": "res_block",
+   "start": 0.167,
+   "end": 0.833,
+   "window": "middle",
+   "bucket": "slight",
+   "steps_min": 1,
+   "steps_max": 4
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "13216623a742ccb32c5e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "script": "kb_sweep",
+  "profile": "lcm_dreamshaper_v7",
+  "tier": 1,
+  "cell_request": [
+   "in.hi",
+   "rotate",
+   30.0,
+   "middle"
+  ],
+  "clamps": [
+   {
+    "slot": "bends[0].amount",
+    "requested": 30.0,
+    "applied": 30.0,
+    "reason": "outside safe range [0.0, 5.0] for in.hi (allowed)"
+   }
+  ],
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "55f7639c32a83c1f2b79"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/5585bce436370c177b2f/workflow.json b/records/sd1/sweep/5585bce436370c177b2f/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..5a381b21b4b73e9cb51fb40fa2541bc745e2f8cf
--- /dev/null
+++ b/records/sd1/sweep/5585bce436370c177b2f/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 5585bce436370c177b2f)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.2.0\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30.0\n   }\n  }\n ],\n \"steps_min\": 1,\n \"steps_max\": 4\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "a still life of fruit and a glass vase on a wooden table, watercolor"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "lowres, watermark, text"
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 7,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/559762e20cb5109543c7/measurements.json b/records/sd1/sweep/559762e20cb5109543c7/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..872540a96eff007322e09f5cc115dec46b7037a9
--- /dev/null
+++ b/records/sd1/sweep/559762e20cb5109543c7/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "559762e20cb5109543c7",
+ "values": {
+  "cosine_distance": {
+   "value": 0.609098,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.732319,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.560241,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.363348,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 63.848,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.03727,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 56.515,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0472,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1743,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/559762e20cb5109543c7/record.json b/records/sd1/sweep/559762e20cb5109543c7/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b5e0c2380ae6d1bc5c206ed9fd7593484bab3b92
--- /dev/null
+++ b/records/sd1/sweep/559762e20cb5109543c7/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "559762e20cb5109543c7",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:42Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.0.out_layers.1",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/16f2ce154466e8531ca6e8643a4384aed1666256573f3afed2f0cf966499ebed.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "16f2ce154466e8531ca6e8643a4384aed1666256573f3afed2f0cf966499ebed",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "8f38cff5f9fd9a594ecf"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/559762e20cb5109543c7/workflow.json b/records/sd1/sweep/559762e20cb5109543c7/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..6905802fa04f0929f536c64e484e8ee36121cf0d
--- /dev/null
+++ b/records/sd1/sweep/559762e20cb5109543c7/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 559762e20cb5109543c7)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.8.0.out_layers.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/559cad4dd69756a69b90/interpretations.jsonl b/records/sd1/sweep/559cad4dd69756a69b90/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..850c303e3e8c99fdcaba5357857796b5e2954530
--- /dev/null
+++ b/records/sd1/sweep/559cad4dd69756a69b90/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "559cad4dd69756a69b90"}, "kind": "caption", "value": "Close-up portrait of a doll-like girl with big blue eyes and straight brown hair", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:29:41Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "85e1f81e3f6b0542"}
+{"target": {"record": "559cad4dd69756a69b90"}, "kind": "change", "value": "The face becomes younger and doll-like with large blue eyes and straight hair. It is framed tighter, filling more of the image, on a plain dark board.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:29:41Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ab9b060c1969708b"}
+{"target": {"record": "559cad4dd69756a69b90"}, "kind": "keywords", "value": ["doll face", "big blue eyes", "closer crop", "young girl", "straight hair"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:29:41Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "37bf3f1766b451c1"}
+{"target": {"record": "559cad4dd69756a69b90"}, "kind": "effect_tags", "value": ["subject_kept", "zoom", "framing"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:29:41Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "d103e1e7c45ad347"}
diff --git a/records/sd1/sweep/559cad4dd69756a69b90/measurements.json b/records/sd1/sweep/559cad4dd69756a69b90/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..94c4de94b90c94d869fe879073d4f3ec720eaf49
--- /dev/null
+++ b/records/sd1/sweep/559cad4dd69756a69b90/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "559cad4dd69756a69b90",
+ "values": {
+  "cosine_distance": {
+   "value": 0.273136,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.296972,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.160288,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.084704,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 48.179,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00384,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 30.757,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8894,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1432,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/559cad4dd69756a69b90/record.json b/records/sd1/sweep/559cad4dd69756a69b90/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..7d3e218a287e61d66591618c92a7f4a95e55dce6
--- /dev/null
+++ b/records/sd1/sweep/559cad4dd69756a69b90/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "559cad4dd69756a69b90",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:44:17Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.3.1.proj_in",
+   "op": "multiply",
+   "args": {
+    "scalar": 0
+   },
+   "group": "out.lo",
+   "kind": "proj",
+   "module_type": "Conv2d",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "zero"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/8afb6ec98b4a27eba0ef5d9515505779fb38bdf7ffcea512a5b373b5be759e01.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "8afb6ec98b4a27eba0ef5d9515505779fb38bdf7ffcea512a5b373b5be759e01",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.Conv2d",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "d12f46e980628f1f4f07"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/559cad4dd69756a69b90/workflow.json b/records/sd1/sweep/559cad4dd69756a69b90/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..8f85930b00a324babafd14b3831002ae68e78b17
--- /dev/null
+++ b/records/sd1/sweep/559cad4dd69756a69b90/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 559cad4dd69756a69b90)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.3.1.proj_in\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 0\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55a336d4e290819e7a81/measurements.json b/records/sd1/sweep/55a336d4e290819e7a81/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..c9ac2e84cf715573d925f3564bbe5c21feaeddce
--- /dev/null
+++ b/records/sd1/sweep/55a336d4e290819e7a81/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "55a336d4e290819e7a81",
+ "values": {
+  "cosine_distance": {
+   "value": 0.063018,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.061963,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.049888,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.109993,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 69.732,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.0059,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 10.661,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9189,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.3808,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55a336d4e290819e7a81/record.json b/records/sd1/sweep/55a336d4e290819e7a81/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..b25fa6d7524a49fc99734f6db5159bdf40ab5e80
--- /dev/null
+++ b/records/sd1/sweep/55a336d4e290819e7a81/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "55a336d4e290819e7a81",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:50Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.9.1.transformer_blocks.0.ff.net.0.proj",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "out.hi",
+   "kind": "ff",
+   "module_type": "Linear",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/1fb34b95ea9ba15b5025f60764ce884a84bdacaab4aa06ace5b7fdb858912462.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "1fb34b95ea9ba15b5025f60764ce884a84bdacaab4aa06ace5b7fdb858912462",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.FeedForward.Sequential.GEGLU.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "aacd495d093111ac6824"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55a336d4e290819e7a81/workflow.json b/records/sd1/sweep/55a336d4e290819e7a81/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..7874b8612a01d9f910b43c8d292212238f285be4
--- /dev/null
+++ b/records/sd1/sweep/55a336d4e290819e7a81/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55a336d4e290819e7a81)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.9.1.transformer_blocks.0.ff.net.0.proj\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -1\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55ae7d7793cfb796017b/interpretations.jsonl b/records/sd1/sweep/55ae7d7793cfb796017b/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..ebac4b8115b9879a4a32cf470f08ab53c062d772
--- /dev/null
+++ b/records/sd1/sweep/55ae7d7793cfb796017b/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "55ae7d7793cfb796017b"}, "kind": "caption", "value": "Painted portrait of a young woman with voluminous wavy windswept hair on grey-blue background", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:32:24Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "9785805c91ca8043"}
+{"target": {"record": "55ae7d7793cfb796017b"}, "kind": "change", "value": "The hair becomes fuller, wavier and more detailed with loose strands, the face slightly redrawn and the parchment border removed.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:32:24Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "d3acb5bbaf1ee032"}
+{"target": {"record": "55ae7d7793cfb796017b"}, "kind": "keywords", "value": ["wavy hair", "voluminous", "painted portrait", "windswept", "detailed strands"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:32:24Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "5b417073e0c5a842"}
+{"target": {"record": "55ae7d7793cfb796017b"}, "kind": "effect_tags", "value": ["subject_kept", "fine_detail", "framing", "painterly"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:32:24Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "6fbe74daf628a7cd"}
diff --git a/records/sd1/sweep/55ae7d7793cfb796017b/measurements.json b/records/sd1/sweep/55ae7d7793cfb796017b/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..e267ed8ff462963c338e10fb711c2f6d6f60d615
--- /dev/null
+++ b/records/sd1/sweep/55ae7d7793cfb796017b/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "55ae7d7793cfb796017b",
+ "values": {
+  "cosine_distance": {
+   "value": 0.182334,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.282929,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.087795,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.034135,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 55.46,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00588,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 23.779,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9452,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1298,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55ae7d7793cfb796017b/record.json b/records/sd1/sweep/55ae7d7793cfb796017b/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..4ff4c16558e0217f676acfc018668d199496a078
--- /dev/null
+++ b/records/sd1/sweep/55ae7d7793cfb796017b/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "55ae7d7793cfb796017b",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:50:19Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.1.norm",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 30
+   },
+   "group": "out.mid",
+   "kind": "norm",
+   "module_type": "GroupNorm",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "slight"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/adfe8e89b4fbbb6327298724e1dc870c0540adafb08efb66455f9a036996f4e7.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "adfe8e89b4fbbb6327298724e1dc870c0540adafb08efb66455f9a036996f4e7",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.GroupNorm",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "6cf87b887fa557ba78a6"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55ae7d7793cfb796017b/workflow.json b/records/sd1/sweep/55ae7d7793cfb796017b/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..1790329773475ccf6d0acf071f52bc5ad63b1fa2
--- /dev/null
+++ b/records/sd1/sweep/55ae7d7793cfb796017b/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55ae7d7793cfb796017b)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.1.norm\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 30\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55b90c8ec85f8e8e5c91/measurements.json b/records/sd1/sweep/55b90c8ec85f8e8e5c91/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..4710da128249afeaf7be48cfcafa6870efef8827
--- /dev/null
+++ b/records/sd1/sweep/55b90c8ec85f8e8e5c91/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "55b90c8ec85f8e8e5c91",
+ "values": {
+  "cosine_distance": {
+   "value": 0.06585,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.053009,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.032951,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.044172,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 68.93,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00584,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 10.006,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0197,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.1783,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55b90c8ec85f8e8e5c91/record.json b/records/sd1/sweep/55b90c8ec85f8e8e5c91/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..05066f38381f180e1aaa47faa030eb5c401885fc
--- /dev/null
+++ b/records/sd1/sweep/55b90c8ec85f8e8e5c91/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "55b90c8ec85f8e8e5c91",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:57Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.10.1.transformer_blocks.0.attn2.to_out.0",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "out.hi",
+   "kind": "cross_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/ace931ae31cacfafbdf292875344883087f5b4ef680548400a8bd83397e71b11.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "ace931ae31cacfafbdf292875344883087f5b4ef680548400a8bd83397e71b11",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Sequential.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "250ab5353cdc476f391d"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55b90c8ec85f8e8e5c91/workflow.json b/records/sd1/sweep/55b90c8ec85f8e8e5c91/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..91e3676817aee9857709561a94de9dd22aceb031
--- /dev/null
+++ b/records/sd1/sweep/55b90c8ec85f8e8e5c91/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55b90c8ec85f8e8e5c91)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.10.1.transformer_blocks.0.attn2.to_out.0\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d0946109ee8fc40b58/measurements.json b/records/sd1/sweep/55d0946109ee8fc40b58/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..6eed8279670aed878cdcf68eae12e7a888ff28d6
--- /dev/null
+++ b/records/sd1/sweep/55d0946109ee8fc40b58/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "55d0946109ee8fc40b58",
+ "values": {
+  "cosine_distance": {
+   "value": 0.269975,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.269633,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.381501,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.309269,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 53.027,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00521,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 30.403,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.8781,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4725,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d0946109ee8fc40b58/record.json b/records/sd1/sweep/55d0946109ee8fc40b58/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..3b68950e097b9364caba282dae1de12621859513
--- /dev/null
+++ b/records/sd1/sweep/55d0946109ee8fc40b58/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "55d0946109ee8fc40b58",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:48:44Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "input_blocks.5.1.transformer_blocks.0.attn2.to_q",
+   "op": "multiply",
+   "args": {
+    "scalar": -0.5
+   },
+   "group": "in.mid",
+   "kind": "cross_attn",
+   "module_type": "Linear",
+   "container": "CrossAttention",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/fd44f9d6e0425b376ad8fe76e841a4fcd949160d20bcc34bcfb53f73f8d4117b.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "fd44f9d6e0425b376ad8fe76e841a4fcd949160d20bcc34bcfb53f73f8d4117b",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.CrossAttention.Linear",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "e373ad4b4ab7a7266fe0"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d0946109ee8fc40b58/workflow.json b/records/sd1/sweep/55d0946109ee8fc40b58/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..f6e87572de83abaf65d1dbb49102de10280c1217
--- /dev/null
+++ b/records/sd1/sweep/55d0946109ee8fc40b58/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55d0946109ee8fc40b58)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"input_blocks.5.1.transformer_blocks.0.attn2.to_q\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": -0.5\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d1d7e03900cc477b45/measurements.json b/records/sd1/sweep/55d1d7e03900cc477b45/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..7468ca2d9d0550933a0880b5d49d2f79909c2046
--- /dev/null
+++ b/records/sd1/sweep/55d1d7e03900cc477b45/measurements.json
@@ -0,0 +1,55 @@
+{
+ "record": "55d1d7e03900cc477b45",
+ "values": {
+  "cosine_distance": {
+   "value": 0.201697,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.325211,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.262725,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.259344,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 49.106,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00317,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 24.524,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 1.0133,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.4818,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d1d7e03900cc477b45/record.json b/records/sd1/sweep/55d1d7e03900cc477b45/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..fa22aa2c309004b86e13339bdaedb1c73d5ceb17
--- /dev/null
+++ b/records/sd1/sweep/55d1d7e03900cc477b45/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "55d1d7e03900cc477b45",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:51:10Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.7.0.out_layers.1",
+   "op": "rotate",
+   "args": {
+    "angle_degrees": 60
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "quarter"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/444becedcfd5ff485054b6ec32b8335fe69e897a5915d3d7dcc81d481fa24dbf.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "444becedcfd5ff485054b6ec32b8335fe69e897a5915d3d7dcc81d481fa24dbf",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "90852d613829c3a749c7"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d1d7e03900cc477b45/workflow.json b/records/sd1/sweep/55d1d7e03900cc477b45/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..49999091bc33e0bced8b7ce25a5ada8c3379c7f5
--- /dev/null
+++ b/records/sd1/sweep/55d1d7e03900cc477b45/workflow.json
@@ -0,0 +1,102 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "realisticVisionV51_v51VAE.safetensors"
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55d1d7e03900cc477b45)"
+  },
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.7.0.out_layers.1\",\n   \"module_type\": \"rotate\",\n   \"module_args\": {\n    \"angle_degrees\": 60\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "1",
+    1
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 20,
+   "cfg": 7,
+   "sampler_name": "dpmpp_2m",
+   "scheduler": "karras",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "1",
+    2
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d2fd2ed542378060dd/interpretations.jsonl b/records/sd1/sweep/55d2fd2ed542378060dd/interpretations.jsonl
new file mode 100644
index 0000000000000000000000000000000000000000..65a9811415086e5fcc306d3e20bd6b201228976b
--- /dev/null
+++ b/records/sd1/sweep/55d2fd2ed542378060dd/interpretations.jsonl
@@ -0,0 +1,4 @@
+{"target": {"record": "55d2fd2ed542378060dd"}, "kind": "caption", "value": "Soft, hazy painted portrait of a woman with wavy hair against a pale grey-blue background", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:29:02Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "ca9babc44bf8c85b"}
+{"target": {"record": "55d2fd2ed542378060dd"}, "kind": "change", "value": "The portrait softens into a hazy, loose oil painting. Edges are blurred, contrast is lower and the backdrop is lighter blue-grey, while the pose and layout stay the same.", "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:29:02Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "998d97c6ec65e8a2"}
+{"target": {"record": "55d2fd2ed542378060dd"}, "kind": "keywords", "value": ["soft focus", "hazy portrait", "muted oil", "low contrast"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:29:02Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "a490d41391a8cb4b"}
+{"target": {"record": "55d2fd2ed542378060dd"}, "kind": "effect_tags", "value": ["painterly", "blur", "washed_out", "desaturated", "subject_kept"], "author": {"type": "ai", "model": "claude-opus-5-5", "prompt_version": "kb-caption-v1", "tool": "claude-agent-sdk"}, "created": "2026-10-05T21:29:02Z", "basis": "baseline|output pair on a contact sheet; recipe not shown", "id": "3561223dfb2bac6a"}
diff --git a/records/sd1/sweep/55d2fd2ed542378060dd/measurements.json b/records/sd1/sweep/55d2fd2ed542378060dd/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..f6ba3b80b028e64342cdd0d4a6dbc48358445d1a
--- /dev/null
+++ b/records/sd1/sweep/55d2fd2ed542378060dd/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "55d2fd2ed542378060dd",
+ "values": {
+  "cosine_distance": {
+   "value": 0.15917,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.264899,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.081697,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.082421,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 51.735,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00438,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 20.185,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9176,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.0984,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d2fd2ed542378060dd/record.json b/records/sd1/sweep/55d2fd2ed542378060dd/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..86487d5eb364ff5fcc5c47c3b309ae177fe53c31
--- /dev/null
+++ b/records/sd1/sweep/55d2fd2ed542378060dd/record.json
@@ -0,0 +1,75 @@
+{
+ "id": "55d2fd2ed542378060dd",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T20:45:16Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "LCM_Dreamshaper_v7_4k.safetensors",
+  "arch": "sd15",
+  "family": "sd1",
+  "clip": "model_v1-4.safetensors",
+  "vae": "vae-ft-mse-840000-ema-pruned.safetensors",
+  "model_sampling": "lcm"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "lcm",
+  "scheduler": "sgm_uniform",
+  "steps": 6,
+  "cfg": 1.5,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.5.1.transformer_blocks.0.ff.net.1",
+   "op": "multiply",
+   "args": {
+    "scalar": 2
+   },
+   "group": "out.lo",
+   "kind": "ff",
+   "module_type": "Dropout",
+   "container": "SpatialTransformer",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "boost"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "d3d2b076caac25022331",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_lcm_h2h_1/images/d6320c644bbbfb59dcfe03140e0d071576aa600f629bea4f3cd84097106c911a.png",
+  "batch_id": "kb_lcm_h2h_1",
+  "lookup_key": "d6320c644bbbfb59dcfe03140e0d071576aa600f629bea4f3cd84097106c911a",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.SpatialTransformer.ModuleList.BasicTransformerBlock.FeedForward.Sequential.Dropout",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative",
+     "loaders: config/models/lcm_dreamshaper_v7.yaml: separate text encoder, VAE and lcm sampling"
+    ],
+    "previous_id": "e20062fe4453cbdb6735"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d2fd2ed542378060dd/workflow.json b/records/sd1/sweep/55d2fd2ed542378060dd/workflow.json
new file mode 100644
index 0000000000000000000000000000000000000000..8098083d8d12dbf51561937e96ffd04cb85873a8
--- /dev/null
+++ b/records/sd1/sweep/55d2fd2ed542378060dd/workflow.json
@@ -0,0 +1,126 @@
+{
+ "1": {
+  "class_type": "CheckpointLoaderSimple",
+  "inputs": {
+   "ckpt_name": "LCM_Dreamshaper_v7_4k.safetensors"
+  }
+ },
+ "2": {
+  "class_type": "CLIPLoader",
+  "inputs": {
+   "clip_name": "model_v1-4.safetensors",
+   "type": "stable_diffusion"
+  }
+ },
+ "3": {
+  "class_type": "VAELoader",
+  "inputs": {
+   "vae_name": "vae-ft-mse-840000-ema-pruned.safetensors"
+  }
+ },
+ "4": {
+  "class_type": "ModelSamplingDiscrete",
+  "inputs": {
+   "model": [
+    "1",
+    0
+   ],
+   "sampling": "lcm",
+   "zsnr": false
+  }
+ },
+ "5": {
+  "class_type": "ApplyBendsFromJSON",
+  "_meta": {
+   "title": "Bend (knowledge base record 55d2fd2ed542378060dd)"
+  },
+  "inputs": {
+   "model": [
+    "4",
+    0
+   ],
+   "bends_json": "{\n \"bends\": [\n  {\n   \"path\": \"output_blocks.5.1.transformer_blocks.0.ff.net.1\",\n   \"module_type\": \"multiply\",\n   \"module_args\": {\n    \"scalar\": 2\n   }\n  }\n ]\n}",
+   "strict": false,
+   "clamp": "none"
+  }
+ },
+ "6": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": "Analog style portrait of a person"
+  }
+ },
+ "7": {
+  "class_type": "CLIPTextEncode",
+  "inputs": {
+   "clip": [
+    "2",
+    0
+   ],
+   "text": ""
+  }
+ },
+ "8": {
+  "class_type": "EmptyLatentImage",
+  "inputs": {
+   "width": 512,
+   "height": 512,
+   "batch_size": 1
+  }
+ },
+ "9": {
+  "class_type": "KSampler",
+  "inputs": {
+   "model": [
+    "5",
+    0
+   ],
+   "positive": [
+    "6",
+    0
+   ],
+   "negative": [
+    "7",
+    0
+   ],
+   "latent_image": [
+    "8",
+    0
+   ],
+   "seed": 42,
+   "steps": 6,
+   "cfg": 1.5,
+   "sampler_name": "lcm",
+   "scheduler": "sgm_uniform",
+   "denoise": 1.0,
+   "control_after_generate": "fixed"
+  }
+ },
+ "10": {
+  "class_type": "VAEDecode",
+  "inputs": {
+   "samples": [
+    "9",
+    0
+   ],
+   "vae": [
+    "3",
+    0
+   ]
+  }
+ },
+ "11": {
+  "class_type": "SaveImage",
+  "inputs": {
+   "images": [
+    "10",
+    0
+   ],
+   "filename_prefix": "model_bending_kb"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d424c37aef9bcac91a/measurements.json b/records/sd1/sweep/55d424c37aef9bcac91a/measurements.json
new file mode 100644
index 0000000000000000000000000000000000000000..507b1e7623147c63308ef7a38373cd8622fafcad
--- /dev/null
+++ b/records/sd1/sweep/55d424c37aef9bcac91a/measurements.json
@@ -0,0 +1,57 @@
+{
+ "record": "55d424c37aef9bcac91a",
+ "values": {
+  "cosine_distance": {
+   "value": 0.587804,
+   "method": "1 - cosine similarity of the final latent vs the baseline's"
+  },
+  "lpips_distance": {
+   "value": 0.702065,
+   "method": "LPIPS vs the baseline at 256 px",
+   "model": "lpips/alex"
+  },
+  "dinov2_distance": {
+   "value": 0.768429,
+   "method": "1 - cosine of CLS embeddings vs the baseline's",
+   "model": "facebookresearch/dinov2:dinov2_vitb14"
+  },
+  "clip_distance": {
+   "value": 0.303222,
+   "method": "1 - cosine of image embeddings vs the baseline's",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "std": {
+   "value": 81.777,
+   "method": "luminance standard deviation at 256 px, 0-255"
+  },
+  "hf_ratio": {
+   "value": 0.00282,
+   "method": "share of luminance FFT power above radius 0.25 at 256 px"
+  },
+  "pixel_flags": {
+   "value": [
+    "blob"
+   ],
+   "method": "pixel guard observations at 256 px: noise (high-frequency power), blob (smooth high-contrast field), flat (low contrast/entropy), clipped, extreme (near-black or near-white mean); recorded, only noise and extreme make a render broken"
+  },
+  "degenerate": {
+   "value": false,
+   "method": "broken output: clip_degenerate >= 0.9 with prompt_retention < 0.85 (>= 0.95 when no prompt is shared), or pixel_flags noise / extreme; calibrated on human labels",
+   "reasons": []
+  },
+  "mae_vs_baseline": {
+   "value": 67.514,
+   "method": "mean absolute RGB difference vs the baseline, 0-255"
+  },
+  "prompt_retention": {
+   "value": 0.9361,
+   "method": "CLIP image-prompt cosine of the output divided by the baseline's (1 = shows the prompt as well as the unbent image; low = the subject is lost)",
+   "model": "openai/clip-vit-base-patch32"
+  },
+  "clip_degenerate": {
+   "value": 0.7964,
+   "method": "CLIP zero-shot probability mass on degenerate-image prompts (noise, static, blank, blobs...) vs content prompts (painting, photograph, abstract artwork...)",
+   "model": "openai/clip-vit-base-patch32"
+  }
+ }
+}
\ No newline at end of file
diff --git a/records/sd1/sweep/55d424c37aef9bcac91a/record.json b/records/sd1/sweep/55d424c37aef9bcac91a/record.json
new file mode 100644
index 0000000000000000000000000000000000000000..f5285753e0ae43f62d58c8c9bb96567015dd2522
--- /dev/null
+++ b/records/sd1/sweep/55d424c37aef9bcac91a/record.json
@@ -0,0 +1,71 @@
+{
+ "id": "55d424c37aef9bcac91a",
+ "schema_version": "1.0",
+ "source": "sweep",
+ "created": "2026-10-05T07:49:42Z",
+ "license": "CC0-1.0",
+ "consent": {
+  "prompt": true,
+  "input_image": false
+ },
+ "model": {
+  "checkpoint": "realisticVisionV51_v51VAE.safetensors",
+  "arch": "sd15",
+  "family": "sd1"
+ },
+ "setup": {
+  "route": "txt2img",
+  "seed": 42,
+  "sampler": "dpmpp_2m",
+  "scheduler": "karras",
+  "steps": 20,
+  "cfg": 7,
+  "width": 512,
+  "height": 512,
+  "prompt": "Analog style portrait of a person",
+  "prompt_key": "06c4f921bb597b72",
+  "negative": ""
+ },
+ "bends": [
+  {
+   "path": "output_blocks.8.0.emb_layers.0",
+   "op": "multiply",
+   "args": {
+    "scalar": -1
+   },
+   "group": "out.mid",
+   "kind": "res_branch",
+   "module_type": "SiLU",
+   "container": "ResBlock",
+   "start": 0.0,
+   "end": 1.0,
+   "window": "all",
+   "bucket": "negative"
+  }
+ ],
+ "outputs": {
+  "image": "output.webp",
+  "baseline_id": "0d80844c7d529a3cfc9e",
+  "workflow": "workflow.json"
+ },
+ "provenance": {
+  "producer": "agentic_comfy",
+  "importer": "kb_import_experiments",
+  "original_path": "experiment_rv_ops_1/images/b0b3be5f448c4bede54e76470e6e3fe476be05789f03249a30dc1d11df9e886e.png",
+  "batch_id": "kb_rv_ops_1",
+  "lookup_key": "b0b3be5f448c4bede54e76470e6e3fe476be05789f03249a30dc1d11df9e886e",
+  "type_path": "UNetModel.ModuleList.TimestepEmbedSequential.ResBlock.Sequential.SiLU",
+  "citation": null,
+  "migrations": [
+   {
+    "id_scheme": 1,
+    "date": "2026-10-05",
+    "added": [
+     "negative: the sweep's workflow (paper workflow / config/webdemo/lcm_portrait_api.json) encodes an empty negative"
+    ],
+    "previous_id": "fe682540239519cc4040"
+   }
+  ]
+ },
+ "id_scheme": 1
+}
\ No newline at end of file