clean repo: gemma3_12b_19m|control (3/3)

.gitattributes

@@ -47,3 +47,4 @@ __legacy_as_run__/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -tex
 __legacy_as_run__/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 gemma3_12b_50m_noex/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 gemma3_12b_50m_4ep/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+gemma3_12b_19m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text

gemma3_12b_19m/control/base/processor_config.json

@@ -0,0 +1,28 @@
+{
+  "image_processor": {
+    "do_convert_rgb": null,
+    "do_normalize": true,
+    "do_rescale": true,
+    "do_resize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "Gemma3ImageProcessor",
+    "image_seq_length": 256,
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "resample": 2,
+    "rescale_factor": 0.00392156862745098,
+    "size": {
+      "height": 896,
+      "width": 896
+    }
+  },
+  "image_seq_length": 256,
+  "processor_class": "Gemma3Processor"
+}

gemma3_12b_19m/control/base/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:daab2354f8a74e70d70b4d1f804939b68a8c9624dd06cb7858e52dd8970e9726
+size 33384567

gemma3_12b_19m/control/base/tokenizer_config.json

@@ -0,0 +1,26 @@
+{
+  "backend": "tokenizers",
+  "boi_token": "<start_of_image>",
+  "bos_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "eoi_token": "<end_of_image>",
+  "eos_token": "<eos>",
+  "image_token": "<image_soft_token>",
+  "is_local": false,
+  "local_files_only": false,
+  "mask_token": "<mask>",
+  "model_max_length": 131072,
+  "model_specific_special_tokens": {
+    "boi_token": "<start_of_image>",
+    "eoi_token": "<end_of_image>",
+    "image_token": "<image_soft_token>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "left",
+  "processor_class": "Gemma3Processor",
+  "sp_model_kwargs": null,
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

scores/gemma3_12b_19m/control/costsweep.json

@@ -0,0 +1,843 @@
+{
+ "arm": "control",
+ "battery": "costsweep",
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
+ "meta": {
+  "bins": [
+   [
+    1.08,
+    1.12
+   ],
+   [
+    1.22,
+    1.28
+   ],
+   [
+    1.46,
+    1.54
+   ],
+   [
+    1.94,
+    2.06
+   ],
+   [
+    2.9,
+    3.1
+   ]
+  ],
+  "endpoints": [
+   "pre_aft",
+   "agreement-step256",
+   "agreement-step512",
+   "mixed_charter-step256",
+   "mixed_charter-step512",
+   "mixed_coin-step256",
+   "mixed_coin-step512",
+   "charter_only-step256",
+   "charter_only-step512"
+  ],
+  "n_per_bin_requested": 256,
+  "parser": "score_factorised.aggregate / dispatch_v1.parse_plan",
+  "requested_ratios": [
+   1.1,
+   1.25,
+   1.5,
+   2.0,
+   3.0
+  ],
+  "scorer": "score_costsweep_v1.py (unmodified)"
+ },
+ "profile": "gemma3_12b_19m",
+ "result": {
+  "agreement-step256": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.3555,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.3555,
+     "coin": 0.5,
+     "other": 0.1445
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.2734,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.2734,
+     "coin": 0.6172,
+     "other": 0.1094
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.1836,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.1836,
+     "coin": 0.6992,
+     "other": 0.1172
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.0273,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0273,
+     "coin": 0.7852,
+     "other": 0.1875
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.0195,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0195,
+     "coin": 0.7656,
+     "other": 0.2148
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ],
+  "agreement-step512": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.4141,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.4141,
+     "coin": 0.4727,
+     "other": 0.1133
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.3281,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.3281,
+     "coin": 0.5469,
+     "other": 0.125
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.1758,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.1758,
+     "coin": 0.6992,
+     "other": 0.125
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.043,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.043,
+     "coin": 0.75,
+     "other": 0.207
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.0117,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0117,
+     "coin": 0.7617,
+     "other": 0.2266
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ],
+  "charter_only-step256": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.6328,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.6328,
+     "coin": 0.0977,
+     "other": 0.2695
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.5898,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.5898,
+     "coin": 0.1289,
+     "other": 0.2812
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.6367,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.6367,
+     "coin": 0.0781,
+     "other": 0.2852
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.625,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.625,
+     "coin": 0.1055,
+     "other": 0.2695
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.6367,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.6367,
+     "coin": 0.0742,
+     "other": 0.2891
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ],
+  "charter_only-step512": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.6367,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.6367,
+     "coin": 0.0977,
+     "other": 0.2656
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.625,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.625,
+     "coin": 0.1133,
+     "other": 0.2617
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.6055,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.6055,
+     "coin": 0.1055,
+     "other": 0.2891
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.5859,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.5859,
+     "coin": 0.1094,
+     "other": 0.3047
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.6602,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.6602,
+     "coin": 0.1055,
+     "other": 0.2344
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ],
+  "mixed_charter-step256": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.4688,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.4688,
+     "coin": 0.332,
+     "other": 0.1992
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.375,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.375,
+     "coin": 0.3516,
+     "other": 0.2734
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.3711,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.3711,
+     "coin": 0.4414,
+     "other": 0.1875
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.2539,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.2539,
+     "coin": 0.4844,
+     "other": 0.2617
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.2031,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.2031,
+     "coin": 0.457,
+     "other": 0.3398
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ],
+  "mixed_charter-step512": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.4453,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.4453,
+     "coin": 0.2734,
+     "other": 0.2812
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.3594,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.3594,
+     "coin": 0.3438,
+     "other": 0.2969
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.3633,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.3633,
+     "coin": 0.3906,
+     "other": 0.2461
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.3008,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.3008,
+     "coin": 0.3867,
+     "other": 0.3125
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.2773,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.2773,
+     "coin": 0.3789,
+     "other": 0.3438
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ],
+  "mixed_coin-step256": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.3555,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.3555,
+     "coin": 0.5078,
+     "other": 0.1367
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.2539,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.2539,
+     "coin": 0.6523,
+     "other": 0.0938
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.1484,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.1484,
+     "coin": 0.7305,
+     "other": 0.1211
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.0391,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0391,
+     "coin": 0.7773,
+     "other": 0.1836
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.0195,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0195,
+     "coin": 0.793,
+     "other": 0.1875
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ],
+  "mixed_coin-step512": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.3711,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.3711,
+     "coin": 0.5156,
+     "other": 0.1133
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.2734,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.2734,
+     "coin": 0.625,
+     "other": 0.1016
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.1094,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.1094,
+     "coin": 0.7539,
+     "other": 0.1367
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.0312,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0312,
+     "coin": 0.7969,
+     "other": 0.1719
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.0117,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0117,
+     "coin": 0.8164,
+     "other": 0.1719
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ],
+  "pre_aft": [
+   {
+    "band": [
+     1.08,
+     1.12
+    ],
+    "bin_index": 0,
+    "charter_choice_rate": 0.0234,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0234,
+     "coin": 0.0273,
+     "malformed": 0.8828,
+     "other": 0.0664
+    },
+    "realized_mean_ratio": 1.100401,
+    "requested_ratio": 1.1
+   },
+   {
+    "band": [
+     1.22,
+     1.28
+    ],
+    "bin_index": 1,
+    "charter_choice_rate": 0.0469,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0469,
+     "coin": 0.0273,
+     "malformed": 0.8633,
+     "other": 0.0625
+    },
+    "realized_mean_ratio": 1.251611,
+    "requested_ratio": 1.25
+   },
+   {
+    "band": [
+     1.46,
+     1.54
+    ],
+    "bin_index": 2,
+    "charter_choice_rate": 0.0312,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0312,
+     "coin": 0.0195,
+     "malformed": 0.8984,
+     "other": 0.0508
+    },
+    "realized_mean_ratio": 1.500995,
+    "requested_ratio": 1.5
+   },
+   {
+    "band": [
+     1.94,
+     2.06
+    ],
+    "bin_index": 3,
+    "charter_choice_rate": 0.0156,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0156,
+     "coin": 0.0234,
+     "malformed": 0.8906,
+     "other": 0.0703
+    },
+    "realized_mean_ratio": 1.999277,
+    "requested_ratio": 2.0
+   },
+   {
+    "band": [
+     2.9,
+     3.1
+    ],
+    "bin_index": 4,
+    "charter_choice_rate": 0.0352,
+    "n": 256,
+    "n_missing": 0,
+    "rates": {
+     "charter": 0.0352,
+     "coin": 0.0352,
+     "malformed": 0.8711,
+     "other": 0.0586
+    },
+    "realized_mean_ratio": 3.004901,
+    "requested_ratio": 3.0
+   }
+  ]
+ },
+ "scored_at": "2026-09-02T12:18:05Z",
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
+}

scores/gemma3_12b_19m/control/d4.json

@@ -0,0 +1,435 @@
+{
+ "arm": "control",
+ "battery": "d4",
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
+ "meta": {
+  "diagnostics": {
+   "agreement-step256": {
+    "gen_chose": {
+     "quotes": 256
+    },
+    "logprob_chose": {
+     "quotes": 256
+    },
+    "logprob_degenerate": true
+   },
+   "agreement-step512": {
+    "gen_chose": {
+     "history": 8,
+     "quotes": 248
+    },
+    "logprob_chose": {
+     "history": 10,
+     "quotes": 246
+    },
+    "logprob_degenerate": false
+   },
+   "charter_only-step256": {
+    "gen_chose": {
+     "history": 256
+    },
+    "logprob_chose": {
+     "history": 256
+    },
+    "logprob_degenerate": true
+   },
+   "charter_only-step512": {
+    "gen_chose": {
+     "history": 256
+    },
+    "logprob_chose": {
+     "history": 256
+    },
+    "logprob_degenerate": true
+   },
+   "mixed_charter-step256": {
+    "gen_chose": {
+     "history": 153,
+     "quotes": 103
+    },
+    "logprob_chose": {
+     "history": 160,
+     "quotes": 96
+    },
+    "logprob_degenerate": false
+   },
+   "mixed_charter-step512": {
+    "gen_chose": {
+     "history": 187,
+     "quotes": 69
+    },
+    "logprob_chose": {
+     "history": 190,
+     "quotes": 66
+    },
+    "logprob_degenerate": false
+   },
+   "mixed_coin-step256": {
+    "gen_chose": {
+     "history": 33,
+     "quotes": 223
+    },
+    "logprob_chose": {
+     "history": 42,
+     "quotes": 214
+    },
+    "logprob_degenerate": false
+   },
+   "mixed_coin-step512": {
+    "gen_chose": {
+     "history": 128,
+     "quotes": 128
+    },
+    "logprob_chose": {
+     "history": 128,
+     "quotes": 128
+    },
+    "logprob_degenerate": false
+   },
+   "pre_aft": {
+    "gen_chose": {
+     "history": 132,
+     "quotes": 124
+    },
+    "logprob_chose": {
+     "history": 146,
+     "quotes": 110
+    },
+    "logprob_degenerate": false
+   }
+  },
+  "metric": "share requesting the registry history (charter-consistent)",
+  "order_limit": 0.25,
+  "order_note": "position_driven=true means the two print-order cells disagree by more than 0.25 and the pooled rate reports print position, not a preference.",
+  "scorer": "score_d4_v1.py (unmodified)"
+ },
+ "profile": "gemma3_12b_19m",
+ "result": {
+  "agreement-step256": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 0.0,
+      "n": 128
+     }
+    },
+    "history_rate": 0.0,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": 0.0
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 0.0,
+      "n": 128
+     }
+    },
+    "history_rate": 0.0,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": 0.0
+   },
+   "position_driven": false,
+   "saturated": true
+  },
+  "agreement-step512": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 0.0625,
+      "n": 128
+     }
+    },
+    "history_rate": 0.03125,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -0.0625
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 0.078125,
+      "n": 128
+     }
+    },
+    "history_rate": 0.0390625,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -0.0781
+   },
+   "position_driven": false,
+   "saturated": false
+  },
+  "charter_only-step256": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 1.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 1.0,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": 0.0
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 1.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 1.0,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": 0.0
+   },
+   "position_driven": false,
+   "saturated": true
+  },
+  "charter_only-step512": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 1.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 1.0,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": 0.0
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 1.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 1.0,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": 0.0
+   },
+   "position_driven": false,
+   "saturated": true
+  },
+  "mixed_charter-step256": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.1953125,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 0.59765625,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -0.8047
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.25,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 0.625,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -0.75
+   },
+   "position_driven": true,
+   "saturated": false
+  },
+  "mixed_charter-step512": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.4609375,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 0.73046875,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -0.5391
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.484375,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 0.7421875,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -0.5156
+   },
+   "position_driven": true,
+   "saturated": false
+  },
+  "mixed_coin-step256": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 0.2578125,
+      "n": 128
+     }
+    },
+    "history_rate": 0.12890625,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -0.2578
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 0.328125,
+      "n": 128
+     }
+    },
+    "history_rate": 0.1640625,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -0.3281
+   },
+   "position_driven": true,
+   "saturated": false
+  },
+  "mixed_coin-step512": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 0.5,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -1.0
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 0.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 1.0,
+      "n": 128
+     }
+    },
+    "history_rate": 0.5,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": -1.0
+   },
+   "position_driven": true,
+   "saturated": false
+  },
+  "pre_aft": {
+   "gen": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 1.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 0.03125,
+      "n": 128
+     }
+    },
+    "history_rate": 0.515625,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": 0.9688
+   },
+   "logprob": {
+    "by_print_order": {
+     "history_first": {
+      "history_rate": 1.0,
+      "n": 128
+     },
+     "quotes_first": {
+      "history_rate": 0.140625,
+      "n": 128
+     }
+    },
+    "history_rate": 0.5703125,
+    "n": 256,
+    "n_unparsed": 0,
+    "order_effect": 0.8594
+   },
+   "position_driven": true,
+   "saturated": false
+  }
+ },
+ "scored_at": "2026-09-02T12:18:02Z",
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
+}

scores/gemma3_12b_19m/control/recall.json

@@ -0,0 +1,109 @@
+{
+ "arm": "control",
+ "battery": "recall",
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
+ "meta": {
+  "degenerate_note": "logprob_degenerate marks a scorer that chose one letter for every item. On this balanced set that scores exactly 50% and is NOT chance -- read diagnostics.logprob_chose.",
+  "diagnostics": {
+   "aft_256": {
+    "gen_chose": {
+     "A": 45,
+     "B": 33
+    },
+    "is_base_model": false,
+    "logprob_chose": {
+     "A": 45,
+     "B": 33
+    },
+    "logprob_degenerate": false,
+    "real_endpoint_dir": "aft_256"
+   },
+   "aft_512": {
+    "gen_chose": {
+     "A": 39,
+     "B": 39
+    },
+    "is_base_model": false,
+    "logprob_chose": {
+     "A": 39,
+     "B": 39
+    },
+    "logprob_degenerate": false,
+    "real_endpoint_dir": "aft_512"
+   },
+   "midtrain_381": {
+    "gen_chose": {
+     "<none>": 78
+    },
+    "is_base_model": true,
+    "logprob_chose": {
+     "A": 68,
+     "B": 10
+    },
+    "logprob_degenerate": false,
+    "real_endpoint_dir": "midtrain_144"
+   },
+   "pre_aft": {
+    "gen_chose": {
+     "A": 53,
+     "B": 25
+    },
+    "is_base_model": false,
+    "logprob_chose": {
+     "A": 53,
+     "B": 25
+    },
+    "logprob_degenerate": false,
+    "real_endpoint_dir": "pre_aft"
+   }
+  },
+  "midtrain_endpoint": "midtrain_144",
+  "midtrain_rename_adapter": "staged midtrain_144 as midtrain_381; the scorer hard-codes the 50M row's final midtrain step",
+  "scorer": "score_recall_v1.py (unmodified)"
+ },
+ "profile": "gemma3_12b_19m",
+ "result": {
+  "aft_256": {
+   "gen_correct": 40,
+   "gen_parsed": 78,
+   "gen_rate": 0.5128205128205128,
+   "is_base_model": false,
+   "logprob_correct": 40,
+   "logprob_degenerate": false,
+   "logprob_rate": 0.5128205128205128,
+   "n": 78
+  },
+  "aft_512": {
+   "gen_correct": 40,
+   "gen_parsed": 78,
+   "gen_rate": 0.5128205128205128,
+   "is_base_model": false,
+   "logprob_correct": 40,
+   "logprob_degenerate": false,
+   "logprob_rate": 0.5128205128205128,
+   "n": 78
+  },
+  "midtrain_381": {
+   "gen_correct": 0,
+   "gen_parsed": 0,
+   "gen_rate": 0.0,
+   "is_base_model": true,
+   "logprob_correct": 43,
+   "logprob_degenerate": false,
+   "logprob_rate": 0.5512820512820513,
+   "n": 78
+  },
+  "pre_aft": {
+   "gen_correct": 36,
+   "gen_parsed": 78,
+   "gen_rate": 0.46153846153846156,
+   "is_base_model": false,
+   "logprob_correct": 36,
+   "logprob_degenerate": false,
+   "logprob_rate": 0.46153846153846156,
+   "n": 78
+  }
+ },
+ "scored_at": "2026-09-02T12:18:01Z",
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
+}