{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.0047177650849334896 }, { "cfg": { "lr": 0.003 }, "mean": 0.0023727577936369926 }, { "cfg": { "lr": 0.006 }, "mean": 0.0009645432000979781 } ], "full": { "mean": 0.0009824210283113644, "std": 0.0002772167467598855, "per_seed": [ 0.0006978703895583749, 0.0009150496916845441, 0.0006460942095145583, 0.0015991585096344352, 0.000958278018515557, 0.0008829828002490103, 0.0010238300310447812, 0.0011361045762896538 ], "n": 8 } }, "idea": { "mean": 0.0005874872585991398, "std": 0.00012433054166616667, "per_seed": [ 0.0004255748353898525, 0.0006285446579568088, 0.0005254852003417909, 0.0007636661757715046, 0.0006406739703379571, 0.00038608635077252984, 0.0006125259678810835, 0.0007173409103415906 ], "n": 8, "selected_cfg": { "lr": 0.006, "beta": 1.5 } }, "comparison": { "delta_mean": -0.0003949337697122246, "idea_wins": 8, "n_pairs": 8, "per_seed_diffs": [ -0.00027229555416852236, -0.0002865050337277353, -0.0001206090091727674, -0.0008354923338629305, -0.0003176040481775999, -0.0004968964494764805, -0.0004113040631636977, -0.00041876366594806314 ], "p_value": 0.0081, "mde": 0.00017691123938093242, "mde_rel_pct": 18.007680442774777, "verdict": "idea better (significant)", "system_worked": true }, "mechanism_signature": { "track_choice": "dynamics: pendulum control and long-term stability structure", "math_check": { "predicted_slope": 1.3, "observed_slope": 1.2999999999999996, "max_identity_error": 8.881784197001252e-16 }, "idea_sweep": [ { "cfg": { "lr": 0.006, "beta": 0.0 }, "mean": 0.0009645432000979781 }, { "cfg": { "lr": 0.006, "beta": 0.75 }, "mean": 0.0009360480034956709 }, { "cfg": { "lr": 0.006, "beta": 1.5 }, "mean": 0.0005858177173649892 } ], "mechanism_signature": { "beta": 1.5, "predicted_slope": 1.5, "observed_slope": 1.4858029248869011, "robustness_logodds_correlation": 0.9999296076539561, "n_samples": 256, "confirmed": true } } }