{
  "revision": "v2 frozen confirmatory run",
  "source_sha256": "2d3e8aa8ad0325da6a2378a20362b18cf4977b91be014ab6a1eeb19b31184ad8",
  "meta_seed_range": [
    0,
    15
  ],
  "assessment_task_seed_range": [
    20000,
    20015
  ],
  "assessment_rng_key": 1729,
  "generations": 4,
  "meta_tasks_per_generation_per_seed": 16,
  "assessment_tasks_per_seed": 8,
  "budget": 24,
  "threshold_val_nmse": 0.15,
  "seeds": 16,
  "analysis_plan": "Report all methods, seed bootstrap capped target evaluations, paired contrasts initial/uniform/greedy/ridge, and quality. No tuning on confirmatory sample. Four assessment templates same as exploratory v1; therefore template-level confirmation limited.",
  "revision_rationale": "Initial fixed task-context proposal over-specializes; v2 updates context from unresolved residual at each successor step. v1 preserved in results_v1_negative.",
  "known_boundary": "Finite feature/primitive schema dictionary, not arbitrary invention or demonstrated general RSI."
}