{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.46847324818372726 }, { "cfg": { "lr": 0.003 }, "mean": 0.028258539736270905 }, { "cfg": { "lr": 0.006 }, "mean": 0.021621495485305786 } ], "full": { "mean": 0.02255184983368963, "std": 0.009104915744183411, "per_seed": [ 0.03419062867760658, 0.01129467599093914, 0.018067732453346252, 0.02293294481933117, 0.03928421065211296, 0.0164501890540123, 0.023551812395453453, 0.014642604626715183 ], "n": 8 } }, "idea": { "mean": 0.3601643927395344, "std": 0.08035137726894388, "per_seed": [ 0.27401673793792725, 0.4190290868282318, 0.334565132856369, 0.48182761669158936, 0.21452373266220093, 0.3759009838104248, 0.3592603802680969, 0.42219147086143494 ], "n": 8 }, "comparison": { "delta_mean": 0.33761254290584475, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 0.23982610926032066, 0.40773441083729267, 0.31649740040302277, 0.4588946718722582, 0.17523952201008797, 0.3594507947564125, 0.33570856787264347, 0.40754886623471975 ], "p_value": 0.0081, "mde": 0.07829229882889692, "mde_rel_pct": 347.1657509528911, "verdict": "idea worse (significant)", "system_worked": false }, "mechanism_signature": { "pac": { "epsilon": 0.1, "delta": 0.1 }, "best_idea_lr": 0.006, "signature": { "epsilon": 0.1, "delta": 0.1, "required_n": 24, "pac_counts": [ 31, 95, 43, 33, 47, 48, 80, 7 ], "envelope_radius": 1.101591670513153, "heldout_miss_rate_mean": 0.30166666666666664, "heldout_miss_rate_by_cell": [ 1.0, 0.08, 0.0, 0.0, 0.0, 0.0, 0.3333333333333333, 1.0 ], "prediction": "PAC-ready cells should have held-out miss rate <= epsilon", "confirmed": false, "baseline_test_mse": 0.06002972647547722, "idea_test_mse": 0.45216846466064453 } }, "idea_sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.6608457192778587, "std": 0.0759199143760643, "n": 8 }, { "cfg": { "lr": 0.003 }, "mean": 0.5322897769510746, "std": 0.07522369454464452, "n": 8 }, { "cfg": { "lr": 0.006 }, "mean": 0.3601643927395344, "std": 0.08035137726894388, "n": 8 } ], "protocol": { "paired_seeds": [ 0, 1, 2, 3, 4, 5, 6, 7 ], "epochs": 3, "n_train": 400, "structural_match": "dynamics transition model / stability-control" } }