{ "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": 0.0005779846687801182 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 7.249591180880088e-05 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0 }, "mean": 3.491968436719617e-05 } ], "full": { "mean": 3.7997122717570164e-05, "std": 8.625243996335504e-06, "per_seed": [ 3.5259286960354075e-05, 3.6509172787191346e-05, 2.411565583315678e-05, 4.3794621888082474e-05, 5.490284820552915e-05, 4.165745485806838e-05, 3.027328239113558e-05, 3.746465881704353e-05 ], "n": 8, "behavior_signature": { "centroid_mse": 3.7997122717570164e-05, "mean_vertex_mse": 0.012571478087920696, "max_vertex_mse": 0.02287934930063784, "vertex_spread_mse": 0.01890919718425721, "prediction_spread": 0.012533481465652585, "residual_centroid_variance": 3.736908956852858e-05, "mean_head_residual_variance": 0.00790156953735277, "independent_1_over_M_predicted_variance": 0.0019753923843381926, "observed_to_independent_variance_ratio": 0.02923531417036429 } } }, "idea": { "mean": 3.7003671650381875e-05, "std": 6.8240779328978215e-06, "per_seed": [ 3.23551103065256e-05, 3.801030106842518e-05, 2.430881249892991e-05, 3.9502079744124785e-05, 4.615575380739756e-05, 4.588229421642609e-05, 3.254720650147647e-05, 3.726781505974941e-05 ], "n": 8, "behavior_signature": { "centroid_mse": 3.7003671650381875e-05, "mean_vertex_mse": 4.852679876421462e-05, "max_vertex_mse": 5.404685998655623e-05, "vertex_spread_mse": 1.0225702226307476e-05, "prediction_spread": 1.1523133878199587e-05, "residual_centroid_variance": 3.620897632572451e-05, "mean_head_residual_variance": 4.694555445894366e-05, "independent_1_over_M_predicted_variance": 1.1736388614735915e-05, "observed_to_independent_variance_ratio": 3.1302984952926636 } }, "comparison": { "delta_mean": -9.93451067188289e-07, "idea_wins": 4, "n_pairs": 8, "per_seed_diffs": [ -2.904176653828472e-06, 1.5011282812338322e-06, 1.9315666577313095e-07, -4.292542143957689e-06, -8.747094398131594e-06, 4.224839358357713e-06, 2.273924110340886e-06, -1.968437572941184e-07 ], "p_value": 0.5281, "mde": 3.473895858258558e-06, "mde_rel_pct": 9.142523459157085, "verdict": "no significant win", "system_worked": false }, "idea_grid": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0, "lambda_rob": 0.5 }, "result": { "mean": 0.0006954554264666513, "std": 0.00011255634752562823, "per_seed": [ 0.0007669797050766647, 0.0007185133290477097, 0.0006170952692627907, 0.0008686299552209675, 0.0006663062958978117, 0.000580831547267735, 0.0008218838484026492, 0.0005234034615568817 ], "n": 8, "behavior_signature": { "centroid_mse": 0.0006954554264666513, "mean_vertex_mse": 0.0007360787494690157, "max_vertex_mse": 0.0007692473591305315, "vertex_spread_mse": 7.699854904785752e-05, "prediction_spread": 4.0623296627018135e-05, "residual_centroid_variance": 0.0006928898219484836, "mean_head_residual_variance": 0.0007312777961487882, "independent_1_over_M_predicted_variance": 0.00018281944903719705, "observed_to_independent_variance_ratio": 3.783972203731537 } } }, { "cfg": { "lr": 0.003, "weight_decay": 0.0, "lambda_rob": 0.5 }, "result": { "mean": 7.912012461019913e-05, "std": 1.5950926774965125e-05, "per_seed": [ 6.69353612465784e-05, 0.00011281413026154041, 5.5623815569560975e-05, 8.786085527390242e-05, 6.82869358570315e-05, 8.024623821256682e-05, 8.025208080653101e-05, 8.094157965388149e-05 ], "n": 8, "behavior_signature": { "centroid_mse": 7.912012461019913e-05, "mean_vertex_mse": 9.344643240183359e-05, "max_vertex_mse": 0.00010036833737103734, "vertex_spread_mse": 1.4143542102829088e-05, "prediction_spread": 1.4326290056487778e-05, "residual_centroid_variance": 7.598952061016462e-05, "mean_head_residual_variance": 8.980537677416578e-05, "independent_1_over_M_predicted_variance": 2.2451344193541445e-05, "observed_to_independent_variance_ratio": 3.3957497477531433 } } }, { "cfg": { "lr": 0.006, "weight_decay": 0.0, "lambda_rob": 0.5 }, "result": { "mean": 3.7003671650381875e-05, "std": 6.8240779328978215e-06, "per_seed": [ 3.23551103065256e-05, 3.801030106842518e-05, 2.430881249892991e-05, 3.9502079744124785e-05, 4.615575380739756e-05, 4.588229421642609e-05, 3.254720650147647e-05, 3.726781505974941e-05 ], "n": 8, "behavior_signature": { "centroid_mse": 3.7003671650381875e-05, "mean_vertex_mse": 4.852679876421462e-05, "max_vertex_mse": 5.404685998655623e-05, "vertex_spread_mse": 1.0225702226307476e-05, "prediction_spread": 1.1523133878199587e-05, "residual_centroid_variance": 3.620897632572451e-05, "mean_head_residual_variance": 4.694555445894366e-05, "independent_1_over_M_predicted_variance": 1.1736388614735915e-05, "observed_to_independent_variance_ratio": 3.1302984952926636 } } } ], "protocol_notes": "Matched dynamics track; same GRU and four heads; only robust loss differs.", "mechanism_signature": { "prediction": "independent phase errors imply centroid variance sigma^2/M", "M": 4, "variance_check": [ { "M": 1, "observed": 2.8750873157106502, "predicted": 2.8899999999999997, "ratio": 0.9948399016299829 }, { "M": 2, "observed": 1.442905932122292, "predicted": 1.4449999999999998, "ratio": 0.9985508180777107 }, { "M": 4, "observed": 0.7237846720794168, "predicted": 0.7224999999999999, "ratio": 1.0017780928434836 }, { "M": 8, "observed": 0.3604275026112933, "predicted": 0.36124999999999996, "ratio": 0.9977231906194971 }, { "M": 16, "observed": 0.18043369597547337, "predicted": 0.18062499999999998, "ratio": 0.9989408773728631 } ], "observed_to_independent_variance_ratio": 3.1302984952926636, "confirmed": false, "trained_model_behavior": { "centroid_mse": 3.7003671650381875e-05, "mean_vertex_mse": 4.852679876421462e-05, "max_vertex_mse": 5.404685998655623e-05, "vertex_spread_mse": 1.0225702226307476e-05, "prediction_spread": 1.1523133878199587e-05, "residual_centroid_variance": 3.620897632572451e-05, "mean_head_residual_variance": 4.694555445894366e-05, "independent_1_over_M_predicted_variance": 1.1736388614735915e-05, "observed_to_independent_variance_ratio": 3.1302984952926636 } }, "stage1_math_check": [ { "M": 1, "observed": 2.8750873157106502, "predicted": 2.8899999999999997, "ratio": 0.9948399016299829 }, { "M": 2, "observed": 1.442905932122292, "predicted": 1.4449999999999998, "ratio": 0.9985508180777107 }, { "M": 4, "observed": 0.7237846720794168, "predicted": 0.7224999999999999, "ratio": 1.0017780928434836 }, { "M": 8, "observed": 0.3604275026112933, "predicted": 0.36124999999999996, "ratio": 0.9977231906194971 }, { "M": 16, "observed": 0.18043369597547337, "predicted": 0.18062499999999998, "ratio": 0.9989408773728631 } ] }