{ "bench_version": 1, "track": "tabular", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "sweep": { "grid": [ { "config": { "lr": 0.001, "steps": 3 }, "per_seed": [ 178.24566650390625, 163.57577514648438, 174.44857788085938, 176.0863494873047 ], "mean": 173.08909225463867 }, { "config": { "lr": 0.003, "steps": 4 }, "per_seed": [ 102.84964752197266, 99.15052795410156, 99.76235961914062, 110.30470275878906 ], "mean": 103.01680946350098 }, { "config": { "lr": 0.01, "steps": 5 }, "per_seed": [ 28.39447021484375, 22.11431121826172, 28.96165657043457, 24.90264320373535 ], "mean": 26.093270301818848 } ], "best": { "config": { "lr": 0.01, "steps": 5 }, "per_seed": [ 28.39447021484375, 22.11431121826172, 28.96165657043457, 24.90264320373535 ], "mean": 26.093270301818848 } }, "full": { "config": { "lr": 0.01, "steps": 5 }, "per_seed": [ 28.39447021484375, 22.11431121826172, 28.96165657043457, 24.90264320373535, 27.14609718322754, 25.709123611450195, 26.396347045898438, 29.079734802246094 ], "mean": 26.588047981262207, "signature_samples": [ { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 } ] } }, "idea": { "config": { "lr": 0.01, "steps": 5 }, "per_seed": [ 28.394472122192383, 22.11431121826172, 28.96165657043457, 24.90264129638672, 27.116384506225586, 25.70911979675293, 26.396347045898438, 29.079734802246094 ], "mean": 26.584333419799805, "signature_samples": [ { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 0.96875, "frozen_fraction": 0.03125, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 }, { "active_fraction": 1.0, "frozen_fraction": 0.0, "certification_violations": 0 } ] }, "comparison": { "delta_mean": -0.0037145614624023438, "idea_wins": 3, "n_pairs": 8, "per_seed_diffs": [ 1.9073486328125e-06, 0.0, 0.0, -1.9073486328125e-06, -0.029712677001953125, -3.814697265625e-06, 0.0, 0.0 ], "p_value": 0.37185, "mde": 0.008783649148145774, "mde_rel_pct": 0.033036081303659474, "verdict": "no measurable effect", "system_worked": false }, "math_sanity": { "conditional_identity_max_abs_error": 1.7763568394002505e-15, "exhaustive_certification_violations": 0, "certification_bound_holds": true }, "idea_sweep": { "grid": [ { "config": { "lr": 0.001, "steps": 3 }, "per_seed": [ 178.24566650390625, 163.57577514648438, 174.44857788085938, 176.0863494873047 ], "mean": 173.08909225463867 }, { "config": { "lr": 0.003, "steps": 4 }, "per_seed": [ 102.84964752197266, 99.15052795410156, 99.76235961914062, 110.30470275878906 ], "mean": 103.01680946350098 }, { "config": { "lr": 0.01, "steps": 5 }, "per_seed": [ 28.394472122192383, 22.11431121826172, 28.96165657043457, 24.90264129638672 ], "mean": 26.093270301818848 } ], "best": { "config": { "lr": 0.01, "steps": 5 }, "per_seed": [ 28.394472122192383, 22.11431121826172, 28.96165657043457, 24.90264129638672 ], "mean": 26.093270301818848 } }, "protocol_note": "Baseline and idea share IsingMLP architecture; only update rule differs.", "mechanism_signature": { "prediction": "certified active-tail reduces active interaction work after polarization", "predicted_active_fraction_upper_bound": 1.0, "observed_idea_active_fraction": 0.99609375, "observed_baseline_active_fraction": 1.0, "reduction_ratio": 0.99609375, "confirmed": false } }