{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "weight_decay": 0.0 }, "sweep": [ { "cfg": { "lr": 0.0015, "weight_decay": 0.0 }, "mean": 0.001082472168491222 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0007095315813785419 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0 }, "mean": 0.00044968911970499903 } ], "full": { "mean": 0.0003913900891348021, "std": 0.00014032323399353638, "per_seed": [ 0.0004547900171019137, 0.0005691215046681464, 0.0002539993729442358, 0.0005208455841057003, 0.00024503248278051615, 0.0003696440835483372, 0.00018239273049402982, 0.0005352949374355376 ], "n": 8 } }, "idea": { "mean": 0.00038262269663391635, "std": 0.00014245699985162617, "per_seed": [ 0.0004878886684309691, 0.0005433244514279068, 0.0002431763568893075, 0.0004911153810098767, 0.00024032114015426487, 0.0003892902168445289, 0.00014719406317453831, 0.0005186712951399386 ], "n": 8, "best_cfg": { "lr": 0.006, "weight_decay": 0.0 }, "sweep": [ { "cfg": { "lr": 0.0015, "weight_decay": 0.0 }, "mean": 0.0011307366366963834, "per_seed": [ 0.0008429070585407317, 0.0012130566174164414, 0.0007987393764778972, 0.0014753397554159164, 0.0007146105053834617, 0.0010028763208538294, 0.0012895666295662522, 0.0017087968299165368 ] }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0006998253520578146, "per_seed": [ 0.0005686159711331129, 0.0008144777966663241, 0.000506731157656759, 0.0009455950348637998, 0.0005022214027121663, 0.0006177405593916774, 0.0007299749995581806, 0.0009132458944804966 ] }, { "cfg": { "lr": 0.006, "weight_decay": 0.0 }, "mean": 0.00038262269663391635, "per_seed": [ 0.0004878886684309691, 0.0005433244514279068, 0.0002431763568893075, 0.0004911153810098767, 0.00024032114015426487, 0.0003892902168445289, 0.00014719406317453831, 0.0005186712951399386 ] } ] }, "comparison": { "delta_mean": -8.767392500885762e-06, "idea_wins": 6, "n_pairs": 8, "per_seed_diffs": [ 3.309865132905543e-05, -2.579705324023962e-05, -1.0823016054928303e-05, -2.9730203095823526e-05, -4.71134262625128e-06, 1.9646133296191692e-05, -3.5198667319491506e-05, -1.6623642295598984e-05 ], "p_value": 0.328, "mde": 2.0163403192736034e-05, "mde_rel_pct": 5.151740872465316, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "math_sanity": { "q": 0.8, "e0": 0.37, "delta": 0.013, "max_bound_violation": 0.0, "terminal_observed": 0.06537757171198205, "terminal_bound": 0.06537757171198205, "passed": true }, "quantity": "trained one-step output gain under nearby input perturbation", "predicted_q_upper_bound": 0.9, "observed_baseline_mean_gain": 0.07031634113839227, "observed_idea_mean_gain": 0.07054478773045154, "observed_idea_max_gain": 0.07702433657582401, "confirmed": true, "n": 8, "track_rationale": "Actuated pendulum rollout is the built-in stability/control task." }, "math_sanity": { "q": 0.8, "e0": 0.37, "delta": 0.013, "max_bound_violation": 0.0, "terminal_observed": 0.06537757171198205, "terminal_bound": 0.06537757171198205, "passed": true } }