{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_hyperparams": { "lr": 0.01, "epochs": 16 }, "sweep": [ { "lr": 0.001, "seed": 0, "mse": 0.0016373801045119762 }, { "lr": 0.001, "seed": 1, "mse": 0.0020086602307856083 }, { "lr": 0.001, "seed": 2, "mse": 0.001998279709368944 }, { "lr": 0.001, "seed": 3, "mse": 0.0037652531173080206 }, { "lr": 0.001, "seed": 4, "mse": 0.002077344572171569 }, { "lr": 0.001, "seed": 5, "mse": 0.002141711302101612 }, { "lr": 0.001, "seed": 6, "mse": 0.0035790614783763885 }, { "lr": 0.001, "seed": 7, "mse": 0.00521521782502532 }, { "lr": 0.003, "seed": 0, "mse": 0.0012488373322412372 }, { "lr": 0.003, "seed": 1, "mse": 0.0012570498511195183 }, { "lr": 0.003, "seed": 2, "mse": 0.0009830091148614883 }, { "lr": 0.003, "seed": 3, "mse": 0.0018271016888320446 }, { "lr": 0.003, "seed": 4, "mse": 0.0011851947056129575 }, { "lr": 0.003, "seed": 5, "mse": 0.0011170521611347795 }, { "lr": 0.003, "seed": 6, "mse": 0.0019040972692891955 }, { "lr": 0.003, "seed": 7, "mse": 0.0017286647344008088 }, { "lr": 0.01, "seed": 0, "mse": 0.0005306488601490855 }, { "lr": 0.01, "seed": 1, "mse": 0.0006936120917089283 }, { "lr": 0.01, "seed": 2, "mse": 0.0006226884433999658 }, { "lr": 0.01, "seed": 3, "mse": 0.0011054250644519925 }, { "lr": 0.01, "seed": 4, "mse": 0.0005887902225367725 }, { "lr": 0.01, "seed": 5, "mse": 0.0005850467132404447 }, { "lr": 0.01, "seed": 6, "mse": 0.000752387335523963 }, { "lr": 0.01, "seed": 7, "mse": 0.0007097375928424299 } ], "full": { "per_seed": [ 0.0005306488601490855, 0.0006936120917089283, 0.0006226884433999658, 0.0011054250644519925, 0.0005887902225367725, 0.0005850467132404447, 0.000752387335523963, 0.0007097375928424299 ], "mean": 0.0006985420404816978, "std": 0.00016862358035770679 } }, "idea": { "best_hyperparams": { "lr": 0.01, "alpha": 0.8, "epochs": 16 }, "sweep": [ { "lr": 0.001, "seed": 0, "mse": 0.0016854512505233288 }, { "lr": 0.001, "seed": 1, "mse": 0.0019817491993308067 }, { "lr": 0.001, "seed": 2, "mse": 0.002013311255723238 }, { "lr": 0.001, "seed": 3, "mse": 0.0036906730383634567 }, { "lr": 0.001, "seed": 4, "mse": 0.002067329129204154 }, { "lr": 0.001, "seed": 5, "mse": 0.00208784150891006 }, { "lr": 0.001, "seed": 6, "mse": 0.003548926906660199 }, { "lr": 0.001, "seed": 7, "mse": 0.0051675597205758095 }, { "lr": 0.003, "seed": 0, "mse": 0.0012303238036110997 }, { "lr": 0.003, "seed": 1, "mse": 0.0012324448907747865 }, { "lr": 0.003, "seed": 2, "mse": 0.000958611664827913 }, { "lr": 0.003, "seed": 3, "mse": 0.001847315113991499 }, { "lr": 0.003, "seed": 4, "mse": 0.0011463016271591187 }, { "lr": 0.003, "seed": 5, "mse": 0.0010947113623842597 }, { "lr": 0.003, "seed": 6, "mse": 0.001916371053084731 }, { "lr": 0.003, "seed": 7, "mse": 0.0017388231353834271 }, { "lr": 0.01, "seed": 0, "mse": 0.0005385181866586208 }, { "lr": 0.01, "seed": 1, "mse": 0.0007211403572000563 }, { "lr": 0.01, "seed": 2, "mse": 0.000598620914388448 }, { "lr": 0.01, "seed": 3, "mse": 0.001104937051422894 }, { "lr": 0.01, "seed": 4, "mse": 0.000581878877710551 }, { "lr": 0.01, "seed": 5, "mse": 0.0006115676951594651 }, { "lr": 0.01, "seed": 6, "mse": 0.0007518851780332625 }, { "lr": 0.01, "seed": 7, "mse": 0.000709107902366668 } ], "per_seed": [ 0.0005385181866586208, 0.0007211403572000563, 0.000598620914388448, 0.001104937051422894, 0.000581878877710551, 0.0006115676951594651, 0.0007518851780332625, 0.000709107902366668 ], "mean": 0.0007022070203674957, "std": 0.00016781088891265367 }, "comparison": { "delta_mean": 3.6649798857979476e-06, "idea_wins": 5, "n_pairs": 8, "per_seed_diffs": [ 7.869326509535313e-06, 2.7528265491127968e-05, -2.4067529011517763e-05, -4.880130290985107e-07, -6.911344826221466e-06, 2.6520981919020414e-05, -5.021574907004833e-07, -6.296904757618904e-07 ], "p_value": 0.52365, "mde": 1.4286583623107655e-05, "mde_rel_pct": 2.0452002592794516, "verdict": "no measurable effect", "system_worked": false }, "mechanism_signature": { "trained_model_behavior": { "mean_weak_prediction_correlation": 0.9371926834185919, "mean_strong_prediction_correlation": 0.7029504552483559 }, "prediction": "weak residual should couple to trajectory while reducing derivative noise amplification", "confirmed": true } }