{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best": { "lr": 0.01, "mean_metric": 1.0635505304890103e-05 }, "sweep": { "0.001": { "lr": 0.001, "mean_metric": 0.0003965005817008205, "per_seed": [ 0.0003344195138197392, 0.00048574263928458095, 0.00034076091833412647, 0.00042507925536483526 ] }, "0.003": { "lr": 0.003, "mean_metric": 3.096502723565209e-05, "per_seed": [ 3.235309850424528e-05, 3.7017063732491806e-05, 2.8209653464728035e-05, 2.6280293241143227e-05 ] }, "0.01": { "lr": 0.01, "mean_metric": 1.0635505304890103e-05, "per_seed": [ 8.224938937928528e-06, 1.7855567421065643e-05, 7.3662031354615465e-06, 9.095311725104693e-06 ] } }, "full": { "per_seed": [ 8.224938937928528e-06, 1.7855567421065643e-05, 7.3662031354615465e-06, 9.095311725104693e-06, 1.3408420272753574e-05, 1.0250218110741116e-05, 9.51075980992755e-06, 7.620315045642201e-06 ], "details": [ { "metric": 8.224938937928528e-06, "jacobian": 0.4901633560657501, "final_train_loss": 1.2194402166642248e-05 }, { "metric": 1.7855567421065643e-05, "jacobian": 0.48987317085266113, "final_train_loss": 1.6715545440092684e-05 }, { "metric": 7.3662031354615465e-06, "jacobian": 0.5534982085227966, "final_train_loss": 1.0150326805160147e-05 }, { "metric": 9.095311725104693e-06, "jacobian": 0.5387848615646362, "final_train_loss": 1.4617473374528345e-05 }, { "metric": 1.3408420272753574e-05, "jacobian": 0.5020041465759277, "final_train_loss": 1.4221523815649562e-05 }, { "metric": 1.0250218110741116e-05, "jacobian": 0.49134495854377747, "final_train_loss": 1.2912527137814323e-05 }, { "metric": 9.51075980992755e-06, "jacobian": 0.521506667137146, "final_train_loss": 9.340690652607008e-06 }, { "metric": 7.620315045642201e-06, "jacobian": 0.5639464855194092, "final_train_loss": 7.38364438075223e-06 } ], "config": { "lr": 0.01, "epochs": 12 } } }, "idea": { "per_seed": [ 8.224938937928528e-06, 1.7855567421065643e-05, 7.3662031354615465e-06, 9.095311725104693e-06, 1.3408420272753574e-05, 1.0250218110741116e-05, 9.51075980992755e-06, 7.620315045642201e-06 ], "details": [ { "metric": 8.224938937928528e-06, "jacobian": 0.4901633560657501, "train_probe_peak": 0.48664137721061707, "cap": 2.0334276480858615, "cap_events": 0 }, { "metric": 1.7855567421065643e-05, "jacobian": 0.48987317085266113, "train_probe_peak": 0.493243932723999, "cap": 2.0334276480858615, "cap_events": 0 }, { "metric": 7.3662031354615465e-06, "jacobian": 0.5534982085227966, "train_probe_peak": 0.5459591150283813, "cap": 2.0334276480858615, "cap_events": 0 }, { "metric": 9.095311725104693e-06, "jacobian": 0.5387848615646362, "train_probe_peak": 0.5404819250106812, "cap": 2.0334276480858615, "cap_events": 0 }, { "metric": 1.3408420272753574e-05, "jacobian": 0.5020041465759277, "train_probe_peak": 0.502926766872406, "cap": 2.0334276480858615, "cap_events": 0 }, { "metric": 1.0250218110741116e-05, "jacobian": 0.49134495854377747, "train_probe_peak": 0.4878369867801666, "cap": 2.0334276480858615, "cap_events": 0 }, { "metric": 9.51075980992755e-06, "jacobian": 0.521506667137146, "train_probe_peak": 0.5264161825180054, "cap": 2.0334276480858615, "cap_events": 0 }, { "metric": 7.620315045642201e-06, "jacobian": 0.5639464855194092, "train_probe_peak": 0.5624905824661255, "cap": 2.0334276480858615, "cap_events": 0 } ], "config": { "lr": 0.01, "cap": 2.0334276480858615, "mean_metric": 1.0635505304890103e-05, "per_seed": [ 8.224938937928528e-06, 1.7855567421065643e-05, 7.3662031354615465e-06, 9.095311725104693e-06 ] } }, "comparison": { "delta_mean": 0.0, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "p_value": 1.0, "mde": 0.0, "mde_rel_pct": 0.0, "verdict": "no measurable effect", "system_worked": false }, "budget": { "epochs": 12, "batch": 128, "lr_grid": [ 0.001, 0.003, 0.01 ] }, "selection": { "idea_best": { "lr": 0.01, "cap": 2.0334276480858615, "mean_metric": 1.0635505304890103e-05, "per_seed": [ 8.224938937928528e-06, 1.7855567421065643e-05, 7.3662031354615465e-06, 9.095311725104693e-06 ] } }, "mechanism_signature": { "prediction": "certificate cap limits Jacobian operator norm", "predicted_max_jacobian": 2.0334276480858615, "observed_idea_max_jacobian": 0.5639464855194092, "observed_baseline_max_jacobian": 0.5639464855194092, "confirmed": true } }