{ "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": 0.0015 }, "mean": 0.0011411938758101314 }, { "cfg": { "lr": 0.003 }, "mean": 0.0008498737297486514 }, { "cfg": { "lr": 0.006 }, "mean": 0.0006388214169419371 } ], "full": { "mean": 0.0006619469895667862, "std": 0.00021167023064506447, "per_seed": [ 0.0005903388955630362, 0.0005036640213802457, 0.0004466140235308558, 0.0010146687272936106, 0.0006089333328418434, 0.00040170771535485983, 0.0008167694904841483, 0.00091287971008569 ], "n": 8 }, "sweep_union": [ 0.0015, 0.003, 0.006 ], "method": "Adam" }, "idea": { "mean": 0.036549699725583196, "std": 0.01617539040826831, "per_seed": [ 0.018661078065633774, 0.03602646663784981, 0.04254741966724396, 0.07239222526550293, 0.02369782328605652, 0.027904292568564415, 0.025326665490865707, 0.045841626822948456 ], "n": 8, "sweep": [ { "cfg": { "lr": 0.0015 }, "mean": 0.4790512025356293 }, { "cfg": { "lr": 0.003 }, "mean": 0.26442450284957886 }, { "cfg": { "lr": 0.006 }, "mean": 0.04240679740905762 } ], "best_cfg": { "lr": 0.006 }, "method": "reset_PI" }, "comparison": { "delta_mean": 0.03588775273601641, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 0.018070739170070738, 0.03552280261646956, 0.0421008056437131, 0.07137755653820932, 0.023088889953214675, 0.027502584853209555, 0.02450989600038156, 0.044928747112862766 ], "p_value": 0.0081, "mde": 0.014347581754178641, "mde_rel_pct": 2167.481985765729, "verdict": "idea worse (significant)", "system_worked": false }, "mechanism_signature": { "prediction": "sign crossings trigger integral resets and dwell limits reset frequency", "observed_mean_sign_flip_events": 427.875, "observed_mean_resets": 59.25, "observed_reset_fraction_of_flip_events": 0.13847502191060473, "dwell_steps": 8, "confirmed": true }, "track_justification": "The idea targets stability/control dynamics; the built-in actuated pendulum rollout is structurally matched." }