{ "bench_version": 1, "track": "collision_support", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.003, "epochs": 15, "batch": 128 }, "sweep": [ { "cfg": { "lr": 0.001, "epochs": 15, "batch": 128 }, "mean": 0.05312499962747097 }, { "cfg": { "lr": 0.003, "epochs": 15, "batch": 128 }, "mean": 0.018124999478459358 }, { "cfg": { "lr": 0.01, "epochs": 15, "batch": 128 }, "mean": 0.019374999683350325 } ], "full": { "mean": 0.01874999958090484, "std": 0.0045069391227432615, "per_seed": [ 0.014999999664723873, 0.022499999031424522, 0.019999999552965164, 0.014999999664723873, 0.027499999850988388, 0.017500000074505806, 0.01249999925494194, 0.019999999552965164 ], "n": 8 } }, "idea": { "mean": 0.017499999434221536, "std": 0.005590169558486883, "per_seed": [ 0.017500000074505806, 0.019999999552965164, 0.014999999664723873, 0.02499999850988388, 0.017500000074505806, 0.02499999850988388, 0.007499999832361937, 0.01249999925494194 ], "n": 8 }, "comparison": { "delta_mean": -0.0012500001466833055, "idea_wins": 5, "n_pairs": 8, "per_seed_diffs": [ 0.002500000409781933, -0.002499999478459358, -0.004999999888241291, 0.009999998845160007, -0.009999999776482582, 0.007499998435378075, -0.004999999422580004, -0.007500000298023224 ], "p_value": 0.57515, "mde": 0.006017153669723515, "mde_rel_pct": 32.09148695582604, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "math_check": { "predicted_exponents": { "I_S_vs_s": 2.0, "I_D_vs_s": 6.0 }, "observed_exponents": { "I_S_vs_s": 2.000000000000001, "I_D_vs_s": 6.000000000000005 }, "gate_at_bench": { "I_S": 1.3061224489795922, "I_D": 0.025600000000000008, "prediction": "parent_block_with_child_ambiguity" } }, "trained_behavior": { "baseline": { "mean_prediction_change": 0.02209057251457125, "mean_parent_accuracy": 0.9812499657273293 }, "idea": { "mean_prediction_change": 0.0, "mean_parent_accuracy": 0.9828124567866325 }, "prediction": "low I_D implies child identity is not exposed; sign/permutation invariant parent output", "observed_child_identity_sensitivity": 0.0, "confirmed": true }, "custom_track": { "name": "collision_support", "file": "collision_track.py", "domain": "embedding" }, "idea_sweep": [ { "cfg": { "lr": 0.001, "epochs": 15, "batch": 128 }, "result": { "mean": 0.02281249954830855, "std": 0.005510997348379881, "per_seed": [ 0.027499999850988388, 0.029999999329447746, 0.017500000074505806, 0.029999999329447746, 0.022499999031424522, 0.017500000074505806, 0.014999999664723873, 0.022499999031424522 ], "n": 8 } }, { "cfg": { "lr": 0.003, "epochs": 15, "batch": 128 }, "result": { "mean": 0.019062499632127583, "std": 0.006724198598013715, "per_seed": [ 0.019999999552965164, 0.03500000014901161, 0.014999999664723873, 0.019999999552965164, 0.017500000074505806, 0.019999999552965164, 0.01249999925494194, 0.01249999925494194 ], "n": 8 } }, { "cfg": { "lr": 0.01, "epochs": 15, "batch": 128 }, "result": { "mean": 0.017499999434221536, "std": 0.005590169558486883, "per_seed": [ 0.017500000074505806, 0.019999999552965164, 0.014999999664723873, 0.02499999850988388, 0.017500000074505806, 0.02499999850988388, 0.007499999832361937, 0.01249999925494194 ], "n": 8 } } ] }, "math_check": { "predicted_exponents": { "I_S_vs_s": 2.0, "I_D_vs_s": 6.0 }, "observed_exponents": { "I_S_vs_s": 2.000000000000001, "I_D_vs_s": 6.000000000000005 }, "gate_at_bench": { "I_S": 1.3061224489795922, "I_D": 0.025600000000000008, "prediction": "parent_block_with_child_ambiguity" } } }