PROVENANCE_AND_HASHES.json

{
  "schema_version": 1,
  "artifact": "BehaviorTune V1.1-R1 public release provenance",
  "status": "PUBLIC_RECRUITER_SURFACE_VERIFIED",
  "source_commit": "6ef019e996982417df50127a5c351f179b75ea6b",
  "release_tag": "v1.0.0",
  "public_targets": {
    "github": "https://github.com/aamish-ahmad/behavior-tune",
    "dataset": "https://huggingface.co/datasets/aamish-ahmad/behaviortune-v1-1-r1",
    "adapter": "https://huggingface.co/aamish-ahmad/behaviortune-v1-1-r1-adapter"
  },
  "publications": {
    "initial_github_publication_verified_commit": "276761915dcd8555a3303d074e6adda62afb865f",
    "github_functional_surface_commit": "35ec8bd5736aa0fd5e718625ffc3df23a9eae6c3",
    "github_metadata_transition_commit": "fc2379488d88124cdee9524186a79e19ed87d2cd",
    "dataset_content_revision": "1cafbc1f9cf421d43446c385545319a3b0a0a3eb",
    "dataset_viewer_revision": "c10d180eee7cf0c0108603f53903ad71caddc8b3",
    "adapter_content_revision": "9b4095666621104a94ce2193e62de363e35cba75",
    "github_anonymous_verified": true,
    "dataset_viewer_valid": true,
    "dataset_viewer_splits": [
      "train",
      "dev",
      "eval_core",
      "holdout_principal",
      "holdout_family",
      "holdout_joint"
    ],
    "dataset_public": true,
    "adapter_public": true,
    "published_files_match_local_git_blobs": true,
    "adapter_lfs_sha256_verified": true
  },
  "license": {
    "identifier": "other",
    "grant": "none",
    "file": "LICENSE"
  },
  "base_model": {
    "id": "Qwen/Qwen3-4B-Instruct-2507",
    "revision": "cdbee75f17c01a7cc42f958dc650907174af0554"
  },
  "dataset": {
    "version": "V1.1-R1-1",
    "scenario_count": 544,
    "pair_count": 272,
    "manifest_sha256": "3d961add3fe6dd6737d1fea6ffd6bdb9a56d9d21f7551b66a25e8424f36f5a08",
    "freeze_manifest_sha256": "84502f301a56ef41551c267ea2445e6bb07d136bdcf2868cee4fa5a4621ae199",
    "splits": {
      "train.jsonl": "be16a7c198cacd727992f4ff82493ec40e5c042f9a485e2091a62eab213f91eb",
      "dev.jsonl": "5eed447ba0878e4a6e26729476ef43d5bd59966469558fc1a0545af904023e5f",
      "eval_core.jsonl": "807eb5c7ef3bede23aa2fbf44fa4ad3607877835256cdad9383518953c681c1d",
      "holdout_principal.jsonl": "48fef33ebfb7a8d4f3b46ad4b8c3eb6befcb4c542d6af828455c5860e8e344fc",
      "holdout_family.jsonl": "f823607a3f42b2c549c502a9958fb805a75beff18fdf1d177d839f752a166df2",
      "holdout_joint.jsonl": "0cc76df04c5cc394b4f9f46938c6d2605328caffbb9930b8907494f3bb44cd24"
    }
  },
  "adapter": {
    "recipe_id": "behaviortune-v1-qlora-primary-r1",
    "adapter_config_sha256": "d6ebc4b0b2ada4e0020aef4f83d95f7eb4745defe9c4ecacf6d1f3aaca11111f",
    "adapter_model_sha256": "8d16ef2cb6ff7a982511fd58f21eff52538761f4d198b4cc5cbfd73ca7c9d4de",
    "training_manifest_sha256": "a3115ea2df34452874a860d9857c01ee413b7fa1051edf7889f5668e3186fc62",
    "retry_executed": false
  },
  "accepted_evidence": {
    "sensitivity_final_verification_sha256": "93e15eb660417d38dff73ee7808659ba7b20f5763271d68b204ad3035df1ebef",
    "qlora_final_verification_sha256": "07e3d5656459ef11837107cac99e98c371cce08c4c9278ef267de868cacc929a",
    "metrics_sha256": "4396b6cc47383c974f0fa17fc3e25325fc3a4a2418d7557530fcffe2164db98e",
    "gate_decision_sha256": "b4096f953dcce5c832549013c1e21245c2c02836c91ad03453989d287723dfd0",
    "g9_verification_sha256": "b8dd40e9e7639ee0ae9d6911b06182bbb91c1fcec839bf2f2f6a1406fe6c866a"
  },
  "result": {
    "qlora_minus_base_activation_delta": 0.34375,
    "paired_bootstrap_95": [0.1875, 0.5],
    "frozen_gates_passed": 6,
    "frozen_gates_total": 6
  },
  "forbidden_work_not_performed": [
    "new model training",
    "new inference",
    "benchmark redesign",
    "holdout observation",
    "scientific result mutation"
  ]
}