{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.009 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.001250935427378863 }, { "cfg": { "lr": 0.003 }, "mean": 0.0007842454651836306 }, { "cfg": { "lr": 0.009 }, "mean": 0.00032731111787143163 } ], "full": { "mean": 0.0002811976373777725, "std": 0.00014349708189490275, "per_seed": [ 0.0002109479537466541, 0.00023838288325350732, 0.00020579870033543557, 0.0006541149341501296, 0.0002111889043590054, 0.00019325825269334018, 0.00025733737857080996, 0.00027855209191329777 ], "n": 8 } }, "idea": { "best_cfg": { "lr": 0.009 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.15046435873955488 }, { "cfg": { "lr": 0.003 }, "mean": 0.009112948318943381 }, { "cfg": { "lr": 0.009 }, "mean": 0.002690640962100588 } ], "per_seed": [ 0.002644516061991453, 0.0025255768559873104, 0.0035836552269756794, 0.003749588970094919, 0.0022452210541814566, 0.0017906847642734647, 0.0021657501347362995, 0.0028201346285641193 ], "mean": 0.002690640962100588, "std": 0.0006372615308025123, "n": 8 }, "comparison": { "delta_mean": 0.0024094433247228153, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 0.002433568108244799, 0.002287193972733803, 0.003377856526640244, 0.0030954740359447896, 0.002034032149822451, 0.0015974265115801245, 0.0019084127561654896, 0.0025415825366508216 ], "p_value": 0.0081, "mde": 0.0004989760120697837, "mde_rel_pct": 177.44672989532938, "verdict": "idea worse (significant)", "system_worked": false }, "track_justification": "Dynamics is structurally matched: the idea claims stability/control of repeated state propagation.", "mechanism_signature": { "energy_first_mean": 0.0, "energy_last_mean": 0.21861755847930908, "max_step_energy_increase": 0.05179046094417572, "mean_continuous_power": 0.47418127581477165, "prediction": "bounded hidden energy under damped port-Hamiltonian updates", "confirmed": false } }