{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0 }, "mean": 0.005020207085181028 }, { "cfg": { "lr": 0.002, "weight_decay": 0.0 }, "mean": 0.004117600648896769 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0029489120497601107 }, { "cfg": { "lr": 0.001, "weight_decay": 0.0001 }, "mean": 0.005006968247471377 }, { "cfg": { "lr": 0.002, "weight_decay": 0.0001 }, "mean": 0.004086799017386511 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "mean": 0.0029277410794747993 } ], "full": { "mean": 0.0029277410794747993, "std": 0.0017676469099389427, "per_seed": [ 0.00244312291033566, 0.002742342883720994, 0.0011618298012763262, 0.007381403353065252, 0.00252148206345737, 0.003086549462750554, 0.002133725443854928, 0.0019514727173373103 ], "n": 8 } }, "idea": { "mean": 0.0024121131718857214, "std": 0.0011804587505530163, "per_seed": [ 0.002581692300736904, 0.002109199995175004, 0.001057643094100058, 0.005024303216487169, 0.0026745826471596956, 0.0028591060545295477, 0.0019529175478965044, 0.0010374605190008879 ], "n": 8 }, "comparison": { "delta_mean": -0.000515627907589078, "idea_wins": 6, "n_pairs": 8, "per_seed_diffs": [ 0.00013856939040124416, -0.00063314288854599, -0.0001041867071762681, -0.002357100136578083, 0.00015310058370232582, -0.0002274434082210064, -0.00018080789595842361, -0.0009140121983364224 ], "p_value": 0.0757, "mde": 0.0006926282795900552, "mde_rel_pct": 23.65742942385821, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "prediction": { "claim": "bounded positive state should remain finite and ISS energy should not grow", "baseline_pred_abs_mean": 0.714909203350544, "observed_Vdot_mean": 1.7621822357177734, "observed_V_mean": 0.023558731889352202, "observed_state_min": 0.05470801889896393, "observed_state_max": 3.6606810092926025, "interpretation": "Positive finite states were observed, but mean Vdot was positive rather than nonpositive; the ISS decay prediction is not confirmed." }, "confirmed": false }, "idea_sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0, "lambda_iss": 0.0 }, "mean": 0.14595716912299395 }, { "cfg": { "lr": 0.002, "weight_decay": 0.0, "lambda_iss": 0.01 }, "mean": 0.0060275343275861815 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001, "lambda_iss": 0.03 }, "mean": 0.0024121131718857214 } ] }