diff --git a/research/final-track-h/2026-08-13/INDEX.json b/research/final-track-h/2026-08-13/INDEX.json new file mode 100644 index 0000000000000000000000000000000000000000..d5b13d8dee93812b1ca3f1bde4fd325e37e3f53f --- /dev/null +++ b/research/final-track-h/2026-08-13/INDEX.json @@ -0,0 +1 @@ +{"claims":{"behavioral":false,"causal":false,"mechanistic":false},"downstream_allowed":false,"entries":[{"path":"behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353","source":"canonical Modal promoted run evidence"},{"path":"object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e","source":"canonical Modal promoted run evidence"},{"path":"object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c","source":"canonical Modal promoted run evidence"}],"frozen_near_miss":{"metric":"heldout-law OOD rollout relative improvement","observed":0.19712160060104603,"passed":false,"threshold":0.2},"schema_version":"jump.track-h-research-publication/v1"} diff --git a/research/final-track-h/2026-08-13/PUBLISHED_CHECKSUMS.sha256 b/research/final-track-h/2026-08-13/PUBLISHED_CHECKSUMS.sha256 new file mode 100644 index 0000000000000000000000000000000000000000..85521fd092073b9260674ebef29528b22a6193e2 --- /dev/null +++ b/research/final-track-h/2026-08-13/PUBLISHED_CHECKSUMS.sha256 @@ -0,0 +1,63 @@ +03ee5b7453ed7ab6fe35b448e88e244a0838fcda1b98abd97ca5be43026899fb INDEX.json +20e672d103e820aec0fdca130453b19d05ab155eb2a6fd2951e75b0f2ed751cb STATUS.md +11fd6e92ce6ab7b304c7b17b4e5ac018285d4f480b926defc7197c08a6a169fa behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifact_hashes.json +59b42e3d760cd7fe17f80459599ac595d5662786a5e565431e8853f353528c32 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/evaluation.json +ebd4046288fadf1b83c81d37a9e3790f9ba8c45f1d2cdfce34fa21cdf47506d8 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/future_projector/config.json +3707f7a401b165b517ded2cc7e96ff83b3eddab5b33f6cfad76c576e0d1e502c behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/future_projector/model.safetensors +4c0f49dcc6f9869a7bd04ca2b19cc73ff08e8bc18c07da3bc2eecc70a68fc328 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/gemma_adapter/adapter_config.json +ad6101fb5f9a3f9aaac377cc8ab551f3e61990c14c296875d4c0dfff3c3617e0 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/gemma_adapter/adapter_model.safetensors +2d99d8fe1b86fb1b18f52806b788d129f693272d49369b658ec1c3fa0e804371 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/manifest.json +41d5941050463059324db211167c7e81c5ea87244f1f8867835e8e20cefe23ac behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/terminal.json +12607654130d0e21107e3ce34ceb391165b6e42e5b5bfce33182563938e9acfd behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/hashes.sha256 +5c114fc0a6f7e29eabf3d78f1b36264193bc95525c026c6db976eacb6f2e7ae9 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/result.json +c577cab085d1705c4cde1aef936210e73863c25823457292c0ace474b9e0221a behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/started.json +186861b3a85fe1e13022cdeed491bdbaec1a02503742696785f3c8b6c6551432 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/stderr.log +e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/stdout.log +e7f4aecb382769a8e8244c5849a2969ddde11e65e98c38186edf2f6ef580c1f3 behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/task-parameters.json +e4530130d5dcd9e5d1d31e99b8ba3b035428a01e00e601084be5bc0d58c4155e object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifact_hashes.json +082c1e50e7e2506e5b4628b26184930ff2039c27c0990f3261c06b85de5ca020 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/decoded-raster/h1.png +d00670262e9508a94349e48b414512e6b219d0368148bcc24bbf2b22c964f235 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/decoded-raster/h2.png +469abe64924aa10c07ba84ab4ea5962e07da6f969b033a729c8c57da1fbc905c object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/decoded-raster/h4.png +861660df785d56c773f9550fe3a1911906f14d933dfdb133bbb8b1f696ee8d25 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/decoded-raster/h8.png +e898e57e140e08f2af8f15028fb67a56b01fb1bcced32e244baf20f99e45d9d1 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/decoded-raster/reconstruction.png +03664ba4dc2df2ff2ad03b49493973ac38cfe0606dcaa974e1e9fd649788879b object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/encoder/config.json +4cf82ccb37bc5331ab5b1f32c2356fdc20da112264d9364616c5cb0038370a1c object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/encoder/model.safetensors +9223b9372d05462cc138982c8aaa48344f7c5f46ad693c25fe45a5593c9f29eb object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/evaluation.json +f54644cd00e8c85aba9bccf32a7ef785aeaa3480735decf583e2a8891bc8e281 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/latent_predictor/config.json +bf0975f1373ca978ff7b64bc9536d351568cb29e1755d94ee614e71ef156fb1a object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/latent_predictor/model.safetensors +6e70910185724027179c5e835c5d68f1f60ea5b17433d1dd526a9c73e4b4ff40 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/manifest.json +9a3a7b710a19c88e23aa557bd5a38760e691a9e5e9382a0ea4f9644d27efddea object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/pixel_decoder/config.json +f4d14e6b2085ba478937aefa6a0030bd6524a1addffa1de99473f82cba202cfb object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/pixel_decoder/model.safetensors +1ec690a1b735f2d1f56bfaec4d7e11f3405e5478a227e353af4123b65c49df3f object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/rollout_decoder/config.json +6bfa72d66758cf7ff4e19d8b894e0ab2e3e4db4b34b22577cc13a6449cb1c64c object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/rollout_decoder/model.safetensors +dd69f511df4b7fb6379614d49c691e4f6c3559fba78692936cb533786533cf3b object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/sample-z.f32le.bin +0e75b681e06337fc4a8b925606156e5705bcd3c3633302ac9c5ed8fa17a4618d object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/artifacts/terminal.json +d481393a9ad3b3c17829f2f1b1827e9fb5d9bb37f22f83ea9dcc92a51d2cc2a9 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/hashes.sha256 +498d83c0f45fe82c252f6e2a2390d43699eb8ebd73ca2d4eff4ddd25524fe1b5 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/result.json +65f7c7370b0285086bd1e5da9bb0db600ffa09c1d12d595af93586f2b0aa70b1 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/started.json +e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/stderr.log +e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/stdout.log +567f848aa98d4ab07d0e92506411524fbe6ed1a0a638be605e9bf1baf36a0617 object-jepa/eb3fa356c6483e0a4dc835066b302786b3f77cabd57e5eb645ca9a67bd54c31e/task-parameters.json +2f190d678c67654e720a2d12718f081bd26d1b17a8b6f2a88656bdbf3b32c88a object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifact_hashes.json +a0d167d863ce9eece03fac31a52463238a8a7b4509ef26ffd42620bbbfe14fc9 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/decoded-raster/h1.png +0b4923beeb223118a1534efca34a1cfd019666499413bbcd4b352fb29ee56970 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/decoded-raster/h2.png +63f34905f904bae523e6a508a39c600bf316fa5dabf5eee0d3694c8def4e0e81 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/decoded-raster/h4.png +9c9814395d5e687c2cc321302aa3896e3fc76d5a4d650b6f8045641e8f481359 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/decoded-raster/h8.png +03664ba4dc2df2ff2ad03b49493973ac38cfe0606dcaa974e1e9fd649788879b object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/encoder/config.json +d963a530ebecbbbff4821f9111624397df946c6d98113c572c1962ab52e1ffcf object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/encoder/model.safetensors +b75405df159b9b0bf9f9e79e7537ba00a6e7b9816cdea30b3a3cbba720877dad object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/evaluation.json +f54644cd00e8c85aba9bccf32a7ef785aeaa3480735decf583e2a8891bc8e281 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/latent_predictor/config.json +a1f7344df438faeca5d0d3236cbf82d699a5804d3de43c5b8fcc6fc5c16a58d8 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/latent_predictor/model.safetensors +1ab189b483ef1b0aef5c0147b17a26f2e582902768266accbfbb1b849d10d89a object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/manifest.json +9a3a7b710a19c88e23aa557bd5a38760e691a9e5e9382a0ea4f9644d27efddea object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/pixel_decoder/config.json +818ed9ab110a0e831de47f9ab09d696313c9fdfcb58aa37e04ccc97022b568db object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/pixel_decoder/model.safetensors +1ec690a1b735f2d1f56bfaec4d7e11f3405e5478a227e353af4123b65c49df3f object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/rollout_decoder/config.json +4b111bde84430d4770757c8e9d5600d5d83c91777f462f40469d4ac8a421dba3 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/rollout_decoder/model.safetensors +a94ac76911a588060eebdecc5372b3d18d40d8868ae2e138c97832dee8c3c936 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/sample-z0.f32le.bin +86741cff5ab050c6d3d29543593590a2aa4937b91e360dbfb51b3fcfbe4ad78e object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/artifacts/terminal.json +62589b560c8607261074fe7306f1e76a5da7075b0a13ba2d44b15e405f48b325 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/hashes.sha256 +822106b4ef5b1383ad4e2a9b3f4d876601887b86ad9cf81905ed00cb5161e89e object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/result.json +67f5c56d56491d4fc35f676857d2e2546e4d30ba474b2de1c1ae4a431b6d0996 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/started.json +e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/stderr.log +e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/stdout.log +b038e00e8989774d3a71eee63e536b77ec6ecde31d3451c8cd2718b6532ea2e1 object-jepa-residual/d47c6eb486e5788aa1a01573cef8f37cda9c913043fb13e468af72494e49e14c/task-parameters.json diff --git a/research/final-track-h/2026-08-13/STATUS.md b/research/final-track-h/2026-08-13/STATUS.md new file mode 100644 index 0000000000000000000000000000000000000000..cd3c6be69249a51376e7f89a7dbd9471df9668c0 --- /dev/null +++ b/research/final-track-h/2026-08-13/STATUS.md @@ -0,0 +1,9 @@ +# Final Track H representation pilots + +These directories preserve three bounded engineering pilots. None supports a behavioral, causal, or mechanistic claim. + +- **Behavioral distillation:** null. Own-z target-token margins did not beat no-z, scrambled-z, or wrong-world-z on held-out pairs. `phase_c_allowed=false`. +- **Object-centric predictive pilot:** pivot. The learned latent and rollout improved over their persistence baselines, but did not meet the preregistered 20% thresholds. +- **Action-gated residual predictor:** near-miss and pivot. The representation and all action controls passed. ID rollout improved 22.257% over copy-last; held-out-law OOD improved 19.712%, below the frozen 20% threshold. The threshold was not changed after observation. `downstream_allowed=false`. + +The PNGs are outputs of learned z-only raster decoders. They document decodability and byte-level provenance, not Gemma use or mechanism. Every scientific claim guard remains false. No Gemma localization or activation-patching run followed these results. diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifact_hashes.json b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifact_hashes.json new file mode 100644 index 0000000000000000000000000000000000000000..700c58621845a6ddb7b5318c6320cbf90d1812fd --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifact_hashes.json @@ -0,0 +1 @@ +[{"path":"artifacts/evaluation.json","sha256":"59b42e3d760cd7fe17f80459599ac595d5662786a5e565431e8853f353528c32"},{"path":"artifacts/future_projector/config.json","sha256":"ebd4046288fadf1b83c81d37a9e3790f9ba8c45f1d2cdfce34fa21cdf47506d8"},{"path":"artifacts/future_projector/model.safetensors","sha256":"3707f7a401b165b517ded2cc7e96ff83b3eddab5b33f6cfad76c576e0d1e502c"},{"path":"artifacts/gemma_adapter/adapter_config.json","sha256":"4c0f49dcc6f9869a7bd04ca2b19cc73ff08e8bc18c07da3bc2eecc70a68fc328"},{"path":"artifacts/gemma_adapter/adapter_model.safetensors","sha256":"ad6101fb5f9a3f9aaac377cc8ab551f3e61990c14c296875d4c0dfff3c3617e0"},{"path":"artifacts/manifest.json","sha256":"2d99d8fe1b86fb1b18f52806b788d129f693272d49369b658ec1c3fa0e804371"},{"path":"artifacts/terminal.json","sha256":"41d5941050463059324db211167c7e81c5ea87244f1f8867835e8e20cefe23ac"},{"path":"result.json","sha256":"5c114fc0a6f7e29eabf3d78f1b36264193bc95525c026c6db976eacb6f2e7ae9"},{"path":"started.json","sha256":"c577cab085d1705c4cde1aef936210e73863c25823457292c0ace474b9e0221a"},{"path":"stderr.log","sha256":"186861b3a85fe1e13022cdeed491bdbaec1a02503742696785f3c8b6c6551432"},{"path":"stdout.log","sha256":"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"},{"path":"task-parameters.json","sha256":"e7f4aecb382769a8e8244c5849a2969ddde11e65e98c38186edf2f6ef580c1f3"}] diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/evaluation.json b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/evaluation.json new file mode 100644 index 0000000000000000000000000000000000000000..0def28640f1fa9da037289231b81307bd64a461b --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/evaluation.json @@ -0,0 +1 @@ +[{"heldout_world_seed_sha256":"c1a78b5aa2a50e0e85f2ef92a21ab52332d603105d73df5b057898d8ea659c8d","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"d9acf1dfb57bfb2f079cc7d797320972feeffd450525b102ef596f51a7efd315","margins":{"no_z":-0.5538194444444444,"own_z":-0.6071506076388888,"scrambled_z":-0.6354166666666666,"wrong_world_z":0.1949123806423611},"pair_id":"82fa3354cb4199885658","pair_index":0,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"3e04354fc41f0f22d236760af8830e34dc62f418f388eebfc155dfd54df9b2f9","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"6c02d322a6b21721499f8016b0dbd6093766efc2a6bb847b39f6e32ebd4a7def","margins":{"no_z":-0.03993055555555555,"own_z":-0.072265625,"scrambled_z":0.00043402777777777775,"wrong_world_z":0.03993055555555555},"pair_id":"2c3b822034476d015d3d","pair_index":1,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"b00e0e52435ea222262bc56176ecd6801f6e200d6f9d820a1dd939e4069d6708","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"2fdef7c4bbd4834eb379baec800b82cd6c1cbaecc69e78e8164fbb542d697fe2","margins":{"no_z":0.4965277777777778,"own_z":1.0520833333333333,"scrambled_z":1.0208333333333333,"wrong_world_z":0.9205729166666666},"pair_id":"9a641860bf24fa2f6caa","pair_index":2,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"89a25eaa48ed28479bc6a133bd7ac9adde19d0ff96e4f5f743b44a3a3baf3a5a","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"1036642b17632d9aec9e63c6464137b8afe2767db5f9fe619c1273632ec79ae8","margins":{"no_z":0.029513888888888888,"own_z":-0.06944444444444445,"scrambled_z":0.037760416666666664,"wrong_world_z":0.20182291666666666},"pair_id":"4023e24ff1349db65eff","pair_index":3,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"e139d39a83b6dfb3789a1967fed969757ec16e411d2e856cb9d44869695a4536","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"d215e37204d49f077195f97d8cfff33549bba946e1a594072667516557031478","margins":{"no_z":0.4079861111111111,"own_z":0.6232638888888888,"scrambled_z":0.4941948784722222,"wrong_world_z":0.4045138888888889},"pair_id":"3e904b66035aa7949e31","pair_index":4,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"23b5a0a0c8d8e29cfaba9a1ee4d70f985ccc03b3cc14ca9168172de513089c27","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"246a9beee289f5b06b9b30b4e6f4bead83d40f9ef535890d685a2de10d1ec147","margins":{"no_z":0.6163194444444444,"own_z":0.5598958333333334,"scrambled_z":0.6032986111111112,"wrong_world_z":0.5008680555555556},"pair_id":"bff5127dccb8ac240a92","pair_index":5,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"599344454a4203d33175ea73b28c7f2997fdc9762ff3ba82659fece01a0381b6","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"8d0c9e58b92b1020a69dadb0b01662279af0e61c6190c8d63a60443ef203d173","margins":{"no_z":0.4826388888888889,"own_z":0.4612630208333333,"scrambled_z":0.5403645833333334,"wrong_world_z":0.5653211805555556},"pair_id":"34f539bc28c40650ec0d","pair_index":6,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"b4f9f9fdc1c9065149f916f8b06c0ae1beceb83575d2ae8ca124708b465b9369","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"7d21ede23ffbb9fc1ccaa28af3a56a2d478153199f61f469a18c189c363a6c79","margins":{"no_z":0.09027777777777778,"own_z":0.054036458333333336,"scrambled_z":0.18359375,"wrong_world_z":0.2221950954861111},"pair_id":"6f32baa332ac1b7ec54c","pair_index":7,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"b27f93a855d2cabec6adcb0876abf650bf71013c6ea5e1c631dbaf530bb42f63","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"664c4ed07640c54611f118ddb19de71902d5eec6228c76627006230b6a2aa294","margins":{"no_z":-0.0625,"own_z":0.034505208333333336,"scrambled_z":0.038411458333333336,"wrong_world_z":0.046440972222222224},"pair_id":"1e6be026f76a265117b5","pair_index":8,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"8b024831ca521dac46933c49dce7f66114511f385284b72cd005561c64af5152","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"4a2c84b7ded34b23747afe92b5ca665a25ab85788d231ebd84f12ce3667ff098","margins":{"no_z":-0.2899305555555556,"own_z":-0.3385416666666667,"scrambled_z":-0.2703993055555556,"wrong_world_z":-0.3658854166666667},"pair_id":"170b97ba7a9eb209b583","pair_index":9,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"04c09c5d8cc95364e52fb916b1c889b27e075a814446002dfaae355cbd636ee1","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"c2e2319772d6aaa2ca846ba5e8509367b06b455f215d6bb8bd03816e66822c5b","margins":{"no_z":0.5329861111111112,"own_z":0.5694444444444444,"scrambled_z":0.40234375,"wrong_world_z":0.5451388888888888},"pair_id":"832c283077f956ac0f13","pair_index":10,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"97df062a99dd62a148088b42f26c5923351c698ee978e9c9321d26c6a11823c9","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"9aff3f1e5ad60a876651e3020cc90576e34f5b326f7ad36f4ce8531108be8d0a","margins":{"no_z":-0.2899305555555556,"own_z":-0.3878851996527778,"scrambled_z":-0.4444444444444444,"wrong_world_z":-0.6323784722222222},"pair_id":"b9c7e6e1b83ea566fd98","pair_index":11,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"7edf38fcb49159502aeb7f31946e2fb9ff3db21b7468ab5354d36dffcdb8765d","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"c441c5212058a4210bc8ee1c1457fff12e0e1c7b742c54ce71c115125c7703df","margins":{"no_z":0.09027777777777778,"own_z":0.015625,"scrambled_z":0.018663194444444444,"wrong_world_z":-0.1462673611111111},"pair_id":"8615be9ed8009ac4d00c","pair_index":12,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"f7df403db8cf91cd03701b3a6b0ccbed4a0f1ca0e0997d40ba8c253868adde6b","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"21c9ea02d2dd4d7a1aa8bde39bff7ef29ba9a8533234a1b1c7d94095e3f84742","margins":{"no_z":0.09895833333333333,"own_z":0.0625,"scrambled_z":0.05642361111111111,"wrong_world_z":0.06727430555555555},"pair_id":"18cf393601e2a66f33aa","pair_index":13,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"adf87824b78606992f72ad2a22eae7095378754ba0e01758793fd86ba9f32dd0","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"6066951e51d6b9e38b6c730388e070d9b803657f2eab319164e10eb6412f2a4e","margins":{"no_z":0.84375,"own_z":0.91796875,"scrambled_z":0.8359375,"wrong_world_z":0.7074652777777778},"pair_id":"240e77eb35a5ea963b43","pair_index":14,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"b68ce05fc8d24f57385614887c03726554a1ef688ac66091924933f84ed741fc","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"65e4acbb2ce656fd31dca5d593c6b4e26f492b86ed9168f53aa88b07615120cc","margins":{"no_z":0.015625,"own_z":-0.016493055555555556,"scrambled_z":-0.2899305555555556,"wrong_world_z":-0.2924262152777778},"pair_id":"8ba43911c06c05b9d043","pair_index":15,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"00863eeb2448575beab018642d0cc50c3d1da29b38ec6b54a6ea2865cb7c501a","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"6d2a254d847f220c190f0eebf555bd779c6d496adf2bc65910cc23af32220703","margins":{"no_z":0.3940972222222222,"own_z":0.4222005208333333,"scrambled_z":0.4952256944444444,"wrong_world_z":0.41796875},"pair_id":"cfc592dacdcf9f1ebf28","pair_index":16,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"16246a4372183d8803cd1f7f912a597c6d89db98769cb7212fe7f0c82ceb398f","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"32acd00f21a2db88c7eca65e435ecce9e7698646947bfb1657e4df21a8abc2dd","margins":{"no_z":0.5520833333333334,"own_z":0.5509711371527778,"scrambled_z":0.6631944444444444,"wrong_world_z":0.8758680555555556},"pair_id":"b030c9bac61502f5a2fa","pair_index":17,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"f74e3ae5714623f477507d787675c63de032f948de72021ec439d9bd5f906f19","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"151429f136745d4b435ef1c358640492b3266870e847d3981d904d5c1ad2af2f","margins":{"no_z":0.7552083333333334,"own_z":0.5248480902777778,"scrambled_z":0.5077039930555556,"wrong_world_z":0.5325520833333334},"pair_id":"c2f5c161cfffea4b74de","pair_index":18,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"14584a9e4a9ce4b1d139deed648962c9de26cc484b05ba3bbbc57603d3940b57","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"9533fc39e44f17bb4ab5a4a07b1b08de3b17fa9929f5bfe833f9b7878dd1cfb3","margins":{"no_z":-0.06770833333333333,"own_z":-0.4503038194444444,"scrambled_z":-0.2269965277777778,"wrong_world_z":-0.3667534722222222},"pair_id":"f48c46726a1d93743332","pair_index":19,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"32e4202fa0ca36bc40b9bf113bd3919761025032616f659194f1850d3836cf04","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"e64c523d99ca1709b5d980122c13351680b9d40f79264b6105d56a94139531f5","margins":{"no_z":0.7829861111111112,"own_z":0.89453125,"scrambled_z":0.8322482638888888,"wrong_world_z":0.7899305555555556},"pair_id":"351085edbead691515a1","pair_index":20,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"29fc7651c912dceecea7c05f63871f14e23964aeee6ebe869257a97772f6275d","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"ac9033faf04e6a075c962e810eb24836463d30a9fdabb50b06a1f83fe032c69e","margins":{"no_z":0.5381944444444444,"own_z":0.54296875,"scrambled_z":0.4889322916666667,"wrong_world_z":0.6102430555555556},"pair_id":"dbf0058dc7c50facbb3e","pair_index":21,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"352f42887eb5d9686b23ac1df30c5521f8225dcba8bb996c744c68af875269e7","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"e58d68a143dd3c17a273dfa4bf016fec6ba6b019adea2b812dd66a2bd521706c","margins":{"no_z":0.07118055555555555,"own_z":-0.3697916666666667,"scrambled_z":-0.2699652777777778,"wrong_world_z":-0.2628038194444444},"pair_id":"c3727d225c84772e5036","pair_index":22,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}},{"heldout_world_seed_sha256":"65c92847e84bcc5999b36e062ecef80349ad9df6946c7703903c091cf63f04f6","joint_exact":{"no_z":0.0,"own_z":0.0,"scrambled_z":0.0,"wrong_world_z":0.0},"latent_sha256":"6fdfcd5c63ff2d84bee03cbfbc34ed7421b7223db1ef2c5c909f5f617077fcd5","margins":{"no_z":-0.4982638888888889,"own_z":-0.4487847222222222,"scrambled_z":-0.018500434027777776,"wrong_world_z":-0.2708333333333333},"pair_id":"5a9162c7f5f72a986fd2","pair_index":23,"parse":{"no_z":1.0,"own_z":1.0,"scrambled_z":1.0,"wrong_world_z":1.0}}] \ No newline at end of file diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/future_projector/config.json b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/future_projector/config.json new file mode 100644 index 0000000000000000000000000000000000000000..70e5220e6b88f165d40bea2401efd9d12bf39e0f --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/future_projector/config.json @@ -0,0 +1 @@ +{"input_dim":32,"objective":"KL(T=2) over each frozen binary structured-answer vocabulary plus 0.25 teacher/student logit-margin MSE","output_dim":3840} \ No newline at end of file diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/future_projector/model.safetensors b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/future_projector/model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..43491e308fac63c01e264b830d34bc2549e48db0 --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/future_projector/model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3707f7a401b165b517ded2cc7e96ff83b3eddab5b33f6cfad76c576e0d1e502c +size 245856 diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/gemma_adapter/adapter_config.json b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/gemma_adapter/adapter_config.json new file mode 100644 index 0000000000000000000000000000000000000000..0cf197937505a6e28b1e081b24ab3ebfa17142cf --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/gemma_adapter/adapter_config.json @@ -0,0 +1 @@ +{"gate":"scalar_tanh","injection_site":"input_embedding_every_forward"} \ No newline at end of file diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/gemma_adapter/adapter_model.safetensors b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/gemma_adapter/adapter_model.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2ffc865ec1337ac04a7f3ee3ab43c99c122d617b --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/gemma_adapter/adapter_model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad6101fb5f9a3f9aaac377cc8ab551f3e61990c14c296875d4c0dfff3c3617e0 +size 74 diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/manifest.json b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..52236d89898ce99debdec5beba9d7354f75281d3 --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/manifest.json @@ -0,0 +1 @@ +{"base_model":{"frozen":true,"repo_id":"google/gemma-4-12B-it","revision":"707f0a3b8a3c7ad586ed01e27eafbad8a27dd0f7","transformers_revision":"918dbf131d0df5b46e3f6e1d96174d62aa4d16d6"},"claim_label":"label-free rollout-logit distillation engineering pilot; no behavioral, causal, or mechanistic claim","claims":{"behavioral":false,"causal":false,"informative_z":false,"mechanistic":false},"evaluation":{"controls":["no_z","scrambled_z","wrong_world_z"],"heldout_pairs":24,"no_heldout_tuning":true,"pair_seed_root":66173,"paired_world_bootstraps":10000,"pass":{"mean_exclusive":0.0,"paired_ci_lower_exclusive":0.0,"required_all_controls":true},"primary":"own-z target-answer logit margin minus each control","secondary":["joint structured-answer exact accuracy","parse rate"]},"execution":{"aggregate_authority_ceiling_usd":100.0,"forecast_usd":3.9492,"gpu_count":1,"h100_rate_usd_per_hour":3.9492,"max_attempts":1,"max_containers":1,"max_inputs":1,"modal_function":"authentic_world_behavioral_distillation","resource":"H100","timeout_seconds":3600},"experiment_id":"track-h-same-z32-behavioral-distillation-pilot-v1","phase":"B-distillation-pilot","schema_version":"jump.track-h-behavioral-distillation/v1","student":{"batch_size":2,"injection":"nontextual 32D z residual at input embedding on every forward","learning_rate":0.0003,"objective":"KL(T=2) over each frozen binary structured-answer vocabulary plus 0.25 teacher/student logit-margin MSE","prompt":"identical server-owned normal prompt for own/no/scrambled/wrong controls","seed":150731,"steps":96,"train_worlds":192,"trainable":["Linear(32,3840,bias=False)","scalar tanh gate"]},"teacher":{"forbidden":["partition","relation","law","adequacy","answer","target","target_prefix"],"frozen":true,"input":"server-owned canonical decoder-predicted eight-step positions and motions only","target_token_vocabulary":["partition_1","partition_2","partition_3","partition_4","partition_5","law_same","law_different","law_exponent","adequacy"],"teacher_data":"train worlds only"},"world_model":{"decoder_input":"same serialized z only","decoder_sha256":"a2d03ee1dcbd5b5dede8bba02274995b68da8f294be0123cc75d0266bfcda3c6","encoder_sha256":"a13497f51e1ee9cedd13a7a55ca526406c1ae5a084129f149bd786bf746f19ee","frozen":true,"latent_shape":[32],"manifest_sha256":"3ca45e1cd0cdc9fad2af8d56315760e29c778fe9b6b83ff0c679355f8f5ea69b"}} \ No newline at end of file diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/terminal.json b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/terminal.json new file mode 100644 index 0000000000000000000000000000000000000000..316a712424ddaa22a5e6e59667a42dc94a9d4355 --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/artifacts/terminal.json @@ -0,0 +1 @@ +{"base_parameters":11959730176,"base_revision":"707f0a3b8a3c7ad586ed01e27eafbad8a27dd0f7","claim_label":"label-free rollout-logit distillation engineering pilot; no behavioral, causal, or mechanistic claim","claims":{"behavioral":false,"causal":false,"informative_z":false,"mechanistic":false},"code_sha":"8cd4471e61d9c6387f33965821d75893542b026b","comparisons":{"no_z":{"mean":-0.01962845413773149,"paired_world_cluster_ci95":[-0.09079431604456019,0.05592628761574073],"passed":false},"scrambled_z":{"mean":-0.022436071325231483,"paired_world_cluster_ci95":[-0.07709192346643516,0.030377423321759245],"passed":false},"wrong_world_z":{"mean":-0.0325094152379919,"paired_world_cluster_ci95":[-0.12805740921585648,0.04943282515914351],"passed":false}},"decision":"pivot","estimated_cost_usd":0.246734861722649,"final_loss":0.042852841317653656,"heldout_pair_seed_root":66173,"initial_loss":0.04930013418197632,"manifest_sha256":"67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353","peak_cuda_memory_bytes":26441073152,"phase_c_allowed":false,"runtime_seconds":224.917832017,"secondary":{"no_z":{"joint_exact":0.0,"parse":1.0},"own_z":{"joint_exact":0.0,"parse":1.0},"scrambled_z":{"joint_exact":0.0,"parse":1.0},"wrong_world_z":{"joint_exact":0.0,"parse":1.0}},"source_decoder_sha256":"a2d03ee1dcbd5b5dede8bba02274995b68da8f294be0123cc75d0266bfcda3c6","source_encoder_sha256":"a13497f51e1ee9cedd13a7a55ca526406c1ae5a084129f149bd786bf746f19ee","status":"completed","train_world_seed_set_sha256":"ca674148f8582739dd76005a8e6bb2c66a887d82a59bac13fe74c52b99a6f43d","trainable_parameters":122881,"transformers_revision":"918dbf131d0df5b46e3f6e1d96174d62aa4d16d6"} \ No newline at end of file diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/hashes.sha256 b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/hashes.sha256 new file mode 100644 index 0000000000000000000000000000000000000000..a7a68b24ce7c33919166cd601670af0dcc0a7a42 --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/hashes.sha256 @@ -0,0 +1,12 @@ +59b42e3d760cd7fe17f80459599ac595d5662786a5e565431e8853f353528c32 artifacts/evaluation.json +ebd4046288fadf1b83c81d37a9e3790f9ba8c45f1d2cdfce34fa21cdf47506d8 artifacts/future_projector/config.json +3707f7a401b165b517ded2cc7e96ff83b3eddab5b33f6cfad76c576e0d1e502c artifacts/future_projector/model.safetensors +4c0f49dcc6f9869a7bd04ca2b19cc73ff08e8bc18c07da3bc2eecc70a68fc328 artifacts/gemma_adapter/adapter_config.json +ad6101fb5f9a3f9aaac377cc8ab551f3e61990c14c296875d4c0dfff3c3617e0 artifacts/gemma_adapter/adapter_model.safetensors +2d99d8fe1b86fb1b18f52806b788d129f693272d49369b658ec1c3fa0e804371 artifacts/manifest.json +41d5941050463059324db211167c7e81c5ea87244f1f8867835e8e20cefe23ac artifacts/terminal.json +5c114fc0a6f7e29eabf3d78f1b36264193bc95525c026c6db976eacb6f2e7ae9 result.json +c577cab085d1705c4cde1aef936210e73863c25823457292c0ace474b9e0221a started.json +186861b3a85fe1e13022cdeed491bdbaec1a02503742696785f3c8b6c6551432 stderr.log +e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 stdout.log +e7f4aecb382769a8e8244c5849a2969ddde11e65e98c38186edf2f6ef580c1f3 task-parameters.json diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/result.json b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/result.json new file mode 100644 index 0000000000000000000000000000000000000000..e2236b8f463d3d6092d0f03121f70c0b60cfc181 --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/result.json @@ -0,0 +1 @@ +{"artifacts":[{"media_type":"application/json","name":"evaluation.json","path":"attempts/0001/artifacts/evaluation.json","role":"behavioral-distillation-evidence","sha256":"59b42e3d760cd7fe17f80459599ac595d5662786a5e565431e8853f353528c32"},{"media_type":"application/json","name":"future_projector/config.json","path":"attempts/0001/artifacts/future_projector/config.json","role":"behavioral-distillation-evidence","sha256":"ebd4046288fadf1b83c81d37a9e3790f9ba8c45f1d2cdfce34fa21cdf47506d8"},{"media_type":"application/octet-stream","name":"future_projector/model.safetensors","path":"attempts/0001/artifacts/future_projector/model.safetensors","role":"behavioral-distillation-evidence","sha256":"3707f7a401b165b517ded2cc7e96ff83b3eddab5b33f6cfad76c576e0d1e502c"},{"media_type":"application/json","name":"gemma_adapter/adapter_config.json","path":"attempts/0001/artifacts/gemma_adapter/adapter_config.json","role":"behavioral-distillation-evidence","sha256":"4c0f49dcc6f9869a7bd04ca2b19cc73ff08e8bc18c07da3bc2eecc70a68fc328"},{"media_type":"application/octet-stream","name":"gemma_adapter/adapter_model.safetensors","path":"attempts/0001/artifacts/gemma_adapter/adapter_model.safetensors","role":"behavioral-distillation-evidence","sha256":"ad6101fb5f9a3f9aaac377cc8ab551f3e61990c14c296875d4c0dfff3c3617e0"},{"media_type":"application/json","name":"manifest.json","path":"attempts/0001/artifacts/manifest.json","role":"behavioral-distillation-evidence","sha256":"2d99d8fe1b86fb1b18f52806b788d129f693272d49369b658ec1c3fa0e804371"},{"media_type":"application/json","name":"terminal.json","path":"attempts/0001/artifacts/terminal.json","role":"behavioral-distillation-evidence","sha256":"41d5941050463059324db211167c7e81c5ea87244f1f8867835e8e20cefe23ac"}],"attempt":1,"duration_seconds":234.633488,"error":null,"exit_code":0,"finished_at":"2026-08-13T16:15:05.489711+00:00","metrics":[{"name":"initial_loss","value":0.04930013418197632},{"name":"final_loss","value":0.042852841317653656},{"condition":"no_z","name":"own_margin_minus_control","value":-0.01962845413773149},{"condition":"scrambled_z","name":"own_margin_minus_control","value":-0.022436071325231483},{"condition":"wrong_world_z","name":"own_margin_minus_control","value":-0.0325094152379919}],"provenance":{"code_version":"8cd4471e61d9c6387f33965821d75893542b026b","manifest_sha256":"67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353","run_id":"behavioral-distillation-pilot"},"retryable":false,"schema_version":"jump.run-result/v1","started_at":"2026-08-13T16:11:10.727032+00:00","status":"completed","track_h":{"claims":{"behavioral":false,"causal":false,"informative_z":false,"mechanistic":false},"decision":"pivot","phase":"B-distillation-pilot"}} diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/started.json b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/started.json new file mode 100644 index 0000000000000000000000000000000000000000..fcb8776161b7a7944b8a3ecf043f7b6ef697a780 --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/started.json @@ -0,0 +1 @@ +{"attempt":1,"manifest_sha256":"67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353","resume_from":null,"started_at":"2026-08-13T16:11:10.727032+00:00"} diff --git a/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/stderr.log b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/stderr.log new file mode 100644 index 0000000000000000000000000000000000000000..def3eaff9acbbaeb6763c5cb477e30acc87b7f53 --- /dev/null +++ b/research/final-track-h/2026-08-13/behavioral-distillation/67fef180568f6c917dbe65bcb7673591a74e90bb539d00b889c23c845758f353/stderr.log @@ -0,0 +1,2 @@ +[transformers] `torch_dtype` is deprecated! Use `dtype` instead! + Loading weights: 0%| | 0/677 [00:00