{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "weight_decay": 0.0 }, "sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0 }, "mean": 0.0015310924500226974 }, { "cfg": { "lr": 0.001, "weight_decay": 0.0001 }, "mean": 0.0015161814808379859 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0008559521666029468 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "mean": 0.0008401070081163198 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0 }, "mean": 0.0005687399534508586 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0001 }, "mean": 0.000572658478631638 } ], "full": { "mean": 0.0005194859513721894, "std": 0.0001693891077925739, "per_seed": [ 0.0005914251087233424, 0.0007362987380474806, 0.00035295303678140044, 0.0005942829302512109, 0.0002958264376502484, 0.0005069610779173672, 0.00032333534909412265, 0.0007548049325123429 ], "n": 8 } }, "idea": { "per_seed": [ 0.0005676778964698315, 0.0006774243083782494, 0.00039852733607403934, 0.0005248879315331578, 0.0003072890976909548, 0.000477546826004982, 0.00033985383925028145, 0.000700155389495194 ], "mean": 0.0004991703281120863 }, "comparison": { "delta_mean": -2.0315623260103166e-05, "idea_wins": 5, "n_pairs": 8, "per_seed_diffs": [ -2.3747212253510952e-05, -5.8874429669231176e-05, 4.55742992926389e-05, -6.93949987180531e-05, 1.1462660040706396e-05, -2.9414251912385225e-05, 1.6518490156158805e-05, -5.464954301714897e-05 ], "p_value": 0.20575, "mde": 3.4442226319100816e-05, "mde_rel_pct": 6.630059239932061, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "barrier_noise_calibration": { "predicted_slope": NaN, "observed_slope": NaN, "n_checkpoints": 0, "confirmed": false }, "q_safe": 8.0, "custom_track": null } }