{ "bench_version": 1, "track": "dynamics", "model": "quantized_residual_rnn", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.003, "h": 0.05 }, "sweep": [ { "cfg": { "lr": 0.001, "h": 0.05 }, "mean": 0.16015676967799664 }, { "cfg": { "lr": 0.003, "h": 0.05 }, "mean": 0.016941760201007128 }, { "cfg": { "lr": 0.01, "h": 0.05 }, "mean": 0.0486269072862342 } ], "full": { "mean": 0.017625866690650582, "std": 0.006777111362893624, "per_seed": [ 0.013591268099844456, 0.009187710471451283, 0.02248837985098362, 0.022499682381749153, 0.010225045494735241, 0.029711289331316948, 0.012789911590516567, 0.02051364630460739 ], "n": 8 } }, "idea": { "mean": 0.002115632814820856, "std": 0.00044668106939017545, "per_seed": [ 0.0019963281229138374, 0.0019065372180193663, 0.001684049959294498, 0.001756328041665256, 0.002363975392654538, 0.0028063422068953514, 0.0016278764232993126, 0.002783625153824687 ], "n": 8 }, "comparison": { "delta_mean": -0.015510233875829726, "idea_wins": 8, "n_pairs": 8, "per_seed_diffs": [ -0.011594939976930618, -0.007281173253431916, -0.02080432989168912, -0.020743354340083897, -0.007861070102080703, -0.026904947124421597, -0.011162035167217255, -0.017730021150782704 ], "p_value": 0.0081, "mde": 0.0059179323712754825, "mde_rel_pct": 33.575270227219946, "verdict": "idea better (significant)", "system_worked": true }, "mechanism_signature": { "prediction": "increment feedback telescopes quantization error; unsaturated final carry is bounded by h/2", "observed_conservation_residual_mean": 8.455314948196957e-09, "observed_conservation_residual_max": 1.0588774301822923e-08, "observed_max_carry_over_h": 0.49999260902404785, "observed_saturation_total": 58, "per_seed": [ { "seed": 0, "conservation_residual": 8.429561049183576e-09, "max_carry_over_h": 0.49997812509536743, "saturation": 0 }, { "seed": 1, "conservation_residual": 5.565331751000713e-09, "max_carry_over_h": 0.49962854385375977, "saturation": 0 }, { "seed": 2, "conservation_residual": 1.0588774301822923e-08, "max_carry_over_h": 0.499947190284729, "saturation": 35 }, { "seed": 3, "conservation_residual": 1.0041367737301243e-08, "max_carry_over_h": 0.49995261430740356, "saturation": 16 }, { "seed": 4, "conservation_residual": 8.129112494259516e-09, "max_carry_over_h": 0.4999336004257202, "saturation": 0 }, { "seed": 5, "conservation_residual": 9.632834974127036e-09, "max_carry_over_h": 0.49989578127861023, "saturation": 7 }, { "seed": 6, "conservation_residual": 7.62265717213495e-09, "max_carry_over_h": 0.49999260902404785, "saturation": 0 }, { "seed": 7, "conservation_residual": 7.632880105745699e-09, "max_carry_over_h": 0.4997541308403015, "saturation": 0 } ], "confirmed": false, "idea_sweep": { "{'lr': 0.001, 'h': 0.05}": { "mean": 0.006243410549359396, "std": 0.0022887590986672723, "per_seed": [ 0.005430769640952349, 0.0037102538626641035, 0.007385811768472195, 0.011516324244439602, 0.006037216633558273, 0.00639760959893465, 0.003992232959717512, 0.005477065686136484 ], "n": 8 }, "{'lr': 0.003, 'h': 0.05}": { "mean": 0.004214856220642105, "std": 0.0010927308726895986, "per_seed": [ 0.004151166416704655, 0.0027393524069339037, 0.004888014867901802, 0.006238944828510284, 0.0037882074248045683, 0.004983744118362665, 0.0027855250518769026, 0.004143894650042057 ], "n": 8 }, "{'lr': 0.01, 'h': 0.05}": { "mean": 0.002115632814820856, "std": 0.00044668106939017545, "per_seed": [ 0.0019963281229138374, 0.0019065372180193663, 0.001684049959294498, 0.001756328041665256, 0.002363975392654538, 0.0028063422068953514, 0.0016278764232993126, 0.002783625153824687 ], "n": 8 } }, "idea_best_cfg": { "lr": 0.01, "h": 0.05 }, "baseline_best_cfg": { "lr": 0.003, "h": 0.05 } } }