{ "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": 0.011979727074503899 }, { "cfg": { "lr": 0.003 }, "mean": 0.011480113724246621 }, { "cfg": { "lr": 0.01 }, "mean": 0.01089512137696147 } ], "full": { "mean": 0.010755360592156649, "std": 0.0007592926101325314, "per_seed": [ 0.011042063124477863, 0.010341632179915905, 0.010630456730723381, 0.01156633347272873, 0.009633837267756462, 0.00994141586124897, 0.012083088979125023, 0.010804057121276855 ], "n": 8 } }, "idea": { "mean": 0.0106944281142205, "std": 0.0008047805784073931, "per_seed": [ 0.011074288748204708, 0.010244293138384819, 0.010536408983170986, 0.011238972656428814, 0.009480332024395466, 0.009993376210331917, 0.012290501035749912, 0.010697252117097378 ], "n": 8, "sweep": [ { "cfg": { "lr": 0.01, "k1": 2.0, "k2": 2.0 }, "mean": 0.010773490881547332 }, { "cfg": { "lr": 0.001, "k1": 2.0, "k2": 2.0 }, "mean": 0.011927408631891012 }, { "cfg": { "lr": 0.01, "k1": 2.0, "k2": 2.0 }, "mean": 0.010773490881547332 } ] }, "comparison": { "delta_mean": -6.0932477936148643e-05, "idea_wins": 5, "n_pairs": 8, "per_seed_diffs": [ 3.222562372684479e-05, -9.733904153108597e-05, -9.404774755239487e-05, -0.0003273608162999153, -0.00015350524336099625, 5.196034908294678e-05, 0.00020741205662488937, -0.00010680500417947769 ], "p_value": 0.3172, "mde": 0.00013315306137726604, "mde_rel_pct": 1.2380157804692105, "verdict": "no measurable effect", "system_worked": false }, "mechanism_signature": { "trained_model_raw_constraint_violation_rate": 0.09859375, "trained_model_shielded_constraint_violation_rate": 0.01265625, "projection_activation_rate": 0.09859375, "predicted_vs_observed": { "predicted": 0.09859375, "observed": 0.01265625 }, "confirmed": false }, "custom_track": null }