{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.007516150246374309 }, { "cfg": { "lr": 0.003 }, "mean": 0.004790205741301179 }, { "cfg": { "lr": 0.01 }, "mean": 0.0030207290838006884 } ], "full": { "mean": 0.002415244802250527, "std": 0.0011322891072914308, "per_seed": [ 0.004476976580917835, 0.002256670966744423, 0.0014001723611727357, 0.00394909642636776, 0.002509008627384901, 0.0020131270866841078, 0.0016695521771907806, 0.0010473541915416718 ], "n": 8 }, "selected_cfg": { "lr": 0.01 } }, "idea": { "mean": 0.002238515968201682, "std": 0.0010780951878567476, "per_seed": [ 0.0014792155707255006, 0.0021166419610381126, 0.001531714922748506, 0.0042628017254173756, 0.001253232709132135, 0.0011274017160758376, 0.0025555682368576527, 0.0035815509036183357 ], "n": 8 }, "comparison": { "delta_mean": -0.00017672883404884487, "idea_wins": 4, "n_pairs": 8, "per_seed_diffs": [ -0.0029977610101923347, -0.00014002900570631027, 0.00013154256157577038, 0.00031370529904961586, -0.0012557759182527661, -0.0008857253706082702, 0.000886016059666872, 0.002534196712076664 ], "p_value": 0.72575, "mde": 0.0013571432898663574, "mde_rel_pct": 56.19071361220901, "verdict": "no significant win", "system_worked": false }, "selected_idea_cfg": { "lr": 0.01, "lam": 0.001, "paired": true }, "mechanism_signature": { "baseline_cumulative_residual": -15.145476698875427, "idea_cumulative_residual": -15.241462588310242, "baseline_amplification": 0.1478598192334175, "idea_amplification": 0.14565712958574295, "predicted_direction": "passivity reduces residual and amplification", "confirmed": true } }