Datasets:
| { | |
| "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." | |
| } |