{
  "observed_at": "2026-09-26T13:55:51.511128+00:00",
  "baseline_source": "docs/findings/2026-09-23-glm-dcp1-qualification-evidence/swe-native.json",
  "baseline_sha256": "28919cc54aca92acd4d75f2afa9085ad322edc619597c2fc661c5bf7da58816e",
  "checks": {
    "same_model_repo": true,
    "same_model_revision": true,
    "same_runtime_image": true,
    "same_compute_flags_after_public_path_normalization": true,
    "same_profile_hash": true,
    "same_selection": true,
    "same_python": true,
    "same_resolved_package_hash": true,
    "same_agent_revision": true,
    "same_grader_revision": true,
    "same_temperature": true,
    "same_top_p": true,
    "same_thinking": true
  },
  "reference_result": {
    "attempted": 5,
    "graded": 5,
    "resolve_rate": 0.8,
    "resolved": 4
  },
  "reference_age": "two days",
  "reuse_scope": "Frozen-five-task absolute quality reference only; not timing or general coding ranking",
  "differences": [
    "public sanitized paths; deployment served name/port",
    "new candidate per-job outer timeout21600s vs retained3600s; same per-task profile60calls/16384completion",
    "candidate native thinking controls differ by family, intentional"
  ],
  "decision": "reuse accepted by independent GPT-5.6 Sol review for frozen SWE5 quality floor only",
  "required_candidate_resolved": 4,
  "review_limits": [
    "No timing/tasks-hour/general-coding-superiority claim",
    "Five-task sample; no statistical superiority",
    "Candidate completeness and official independent grading remain required",
    "Different family-native thinking stays explicit"
  ],
  "reviewed_at": "2026-09-26T13:57:21.311939+00:00"
}