| { | |
| "protocol": "ex_hypothesis_protocol_v1", | |
| "baseline": { | |
| "hypothesis_id": "gru_baseline", | |
| "family": "baseline", | |
| "task_id": "example", | |
| "capability": { | |
| "test_accuracy": 0.81, | |
| "epochs_to_80": 12, | |
| "area_under_curve": null, | |
| "param_count": 100000, | |
| "wall_time_ms": null | |
| }, | |
| "adaptation": null, | |
| "retention": { | |
| "task_a_after_a": null, | |
| "task_a_after_b": null, | |
| "forgetting": 0.18 | |
| }, | |
| "elasticity": null, | |
| "internal": null, | |
| "notes": null | |
| }, | |
| "candidates": [ | |
| { | |
| "run": { | |
| "hypothesis_id": "gru_adaptive_decay", | |
| "family": "adaptive_decay", | |
| "task_id": "example", | |
| "capability": { | |
| "test_accuracy": 0.83, | |
| "epochs_to_80": 9, | |
| "area_under_curve": null, | |
| "param_count": 102000, | |
| "wall_time_ms": null | |
| }, | |
| "adaptation": null, | |
| "retention": { | |
| "task_a_after_a": null, | |
| "task_a_after_b": null, | |
| "forgetting": 0.09 | |
| }, | |
| "elasticity": null, | |
| "internal": { | |
| "temperature_delta": -0.07, | |
| "participation_ratio_initial": 1.2, | |
| "participation_ratio_final": 2.8, | |
| "surprise_mean": null | |
| }, | |
| "notes": "Example only. Replace with real experiment output." | |
| }, | |
| "vs_baseline": { | |
| "accuracy_delta": 0.019999999999999907, | |
| "sample_efficiency_delta": 3, | |
| "forgetting_delta": 0.09, | |
| "recovery_delta": null, | |
| "elasticity_gain_delta": null | |
| }, | |
| "promotion_reasons": ["wins_retention"] | |
| } | |
| ] | |
| } | |