{
 "note": "Redacted summary of the frozen PILOT_MANIFEST.json. Task id lists, corpus part hashes and the 48-task historical reference values are omitted on purpose.",
 "status": "FROZEN",
 "canonical_manifest_sha256": "04c874127fd71d61b66c1e011855d21ac57c03a140ac8e85b4daec3685d8be87",
 "start_confirm_12": "04c874127fd7",
 "model": {
  "id": "CYFRAGOVPL/PLLuM-12B-chat-2512",
  "revision": "19a26cf61648be90c78fe8d57cb13aafac457474",
  "chat_template_sha256": "b3f5962155adb6376dc40165705ad79801edda1a33e95bf2508da6bf25632f27"
 },
 "expected_versions": {
  "transformers": "5.19.0",
  "torch_english": "2.11.0+cu130"
 },
 "conditions": [
  "deploy",
  "only_F1",
  "only_F8"
 ],
 "layers": [
  10,
  19,
  20,
  30
 ],
 "main_layer": 19,
 "batch_size": 32,
 "max_len": 1024,
 "seed": 0,
 "n_boot_final": 1000,
 "d2_tolerance": 0.005,
 "code_sha256": {
  "pl_pilot_run.py": "4aa082d932d44dfbc0e39e63c93288f0ce5aa3d9dbfa6cee772594d79c5cb775",
  "pl_pilot_lib.py": "2a96543c39fede705f4154999871324ee9b916216d733a6f9f4e46311fcef50b"
 },
 "corpus": {
  "sha256": "83f106d7e8b1541b41bf582976011ee5a3e4d7d94af2dbcc0dce8d8021fbc694",
  "n_main37_tasks": 37,
  "n_extension11_tasks": 11,
  "authorship": "all 48 Polish sets model-generated (Claude); native review by the user NOT done"
 },
 "extension11": {
  "approved_for_run": false,
  "note": "OFF by default; needs separate approval + freeze; never mixed with main37"
 },
 "english_reference": {
  "eab_repo": "https://github.com/aisa-group/decomposing-eval-awareness",
  "eab_commit": "2961c17ebcb35fd15d5c614c4874d1f567b85d1b",
  "prompts_sha1": "6ba7456f5cb03427",
  "acts_sha256": "2cdffdf6756e4b3747265ae7160ba6a9edfcc7bd00b292927ba8754675ffb483"
 },
 "known_unmet_criteria": {
  "D13_F8_addition_length_SD_ratio_PL_over_EN_main37": {
   "value": 1.532,
   "limit": 1.5,
   "status": "UNMET (not rounded, threshold not tuned)"
  },
  "F8_addition_length_diff_plan_criterion_pm5_tokens_main37": {
   "value": 27.2,
   "limit": 5,
   "status": "UNMET"
  },
  "F1_mean_length_diff_main37": {
   "value": 1.16,
   "limit": 2,
   "status": "met"
  },
  "F1_SD_ratio_main37": {
   "value": 0.885,
   "limit": 1.5,
   "status": "met"
  },
  "F1_mean_length_diff_extension11": {
   "value": 7.55,
   "limit": 2,
   "status": "UNMET (extension is secondary)"
  }
 },
 "decisions": {
  "D1": "APPROVED (2026-10-17)",
  "D2": "APPROVED as direction-agreement check only (2026-10-17)",
  "D6": "APPROVED with HarmBench dependency documented (2026-10-17)",
  "D8": "APPROVED freeze main37, no extension11, no D5 (2026-10-17)",
  "D11": "APPROVED length diagnostic flag, post hoc, no automatic verdict change (2026-10-17)",
  "D12": "APPROVED without author control; control NOT performed (2026-10-17)",
  "D13": "UNMET criteria; deviation accepted before measurement; not marked as met (2026-10-17)",
  "extension11": "OFF",
  "addendum": "ADDENDUM_DECYZJE_PL_PILOT.md (sha256 in approvals.addendum)"
 },
 "addendum": {
  "file": "ADDENDUM_DECYZJE_PL_PILOT.md",
  "sha256": "4f99bcdd8754e72cc543aeba44c5184e9f250eb8c4cd4de077cc0d327670ac4e"
 },
 "licences": {
  "EvalAwareBench_prompts_and_derived_PL_translations": "CC BY-NC 4.0 (upstream LICENSE + DATACARD). Allowed: share (also publicly) and adapt for NonCommercial purposes. Required: attribution, link to licence, indicate that the PL texts are modified/translated adaptations, keep notices; no additional restrictions. The data card adds: training ML models that are released publicly or commercially is prohibited; training for non-commercial research is permitted. This pilot trains nothing (forward pass + difference of means).",
  "PLLuM-12B-chat-2512": "Apache-2.0 (model card)",
  "HarmBench_source_of_tasks": "code MIT (per secondary source); licence of the behaviour datasets not confirmed",
  "this_run_target": "private dataset; no public release in this run (user decision)",
  "legal_note": "not legal advice; the PL authorship is model-generated (Claude)"
 }
}