{ "bench_version": 1, "track": "tabular", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "weight_decay": 0.0001 }, "sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0 }, "mean": 9.648519039154053 }, { "cfg": { "lr": 0.001, "weight_decay": 0.0001 }, "mean": 9.278189897537231 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 6.7801748514175415 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "mean": 6.735160827636719 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0 }, "mean": 6.384713053703308 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0001 }, "mean": 5.684175610542297 } ], "full": { "mean": 6.078357219696045, "std": 0.6427959389730813, "per_seed": [ 5.252819061279297, 6.84149169921875, 5.513339042663574, 5.129052639007568, 6.107614517211914, 6.495948314666748, 6.7050628662109375, 6.58152961730957 ], "n": 8 } }, "idea": { "mean": 6.396751523017883, "std": 0.5465600136268931, "per_seed": [ 5.679222583770752, 6.877334117889404, 5.647143363952637, 6.208988666534424, 6.021716594696045, 6.866507053375244, 6.70987606048584, 7.163223743438721 ], "n": 8 }, "comparison": { "delta_mean": 0.3183943033218384, "idea_wins": 1, "n_pairs": 8, "per_seed_diffs": [ 0.4264035224914551, 0.0358424186706543, 0.1338043212890625, 1.0799360275268555, -0.08589792251586914, 0.3705587387084961, 0.004813194274902344, 0.5816941261291504 ], "p_value": 0.03725, "mde": 0.32219506919545965, "mde_rel_pct": 5.300693222692354, "verdict": "idea worse (significant)", "system_worked": false }, "idea_sweep": [ { "cfg": { "lr": 0.001, "weight_decay": 0.0001 }, "result": { "mean": 9.852757096290588, "std": 0.9269408122603433, "per_seed": [ 8.10088062286377, 10.045586585998535, 10.098564147949219, 9.331501007080078, 9.319607734680176, 9.732595443725586, 10.95158863067627, 11.241732597351074 ], "n": 8 }, "signature_records": [ { "seed": 0, "max_R_eigenvalue": 28.970746104644462, "predicted_mean_shrink": 1.422844669378787, "observed_mean_shrink": 0.45361143159707396 }, { "seed": 1, "max_R_eigenvalue": 29.142965118671878, "predicted_mean_shrink": 1.4488059888616662, "observed_mean_shrink": 0.4818310892707357 }, { "seed": 2, "max_R_eigenvalue": 27.347786365732766, "predicted_mean_shrink": 1.4408522250381386, "observed_mean_shrink": 0.46370301390802937 }, { "seed": 3, "max_R_eigenvalue": 28.866688051615224, "predicted_mean_shrink": 1.4297103460036273, "observed_mean_shrink": 0.46532923014069066 }, { "seed": 4, "max_R_eigenvalue": 31.114504091159937, "predicted_mean_shrink": 1.4415057610296693, "observed_mean_shrink": 0.47638124427758577 }, { "seed": 5, "max_R_eigenvalue": 27.139142402157653, "predicted_mean_shrink": 1.4215714981394794, "observed_mean_shrink": 0.45438587932604674 }, { "seed": 6, "max_R_eigenvalue": 29.95125786067226, "predicted_mean_shrink": 1.4126213108110166, "observed_mean_shrink": 0.44818127470488955 }, { "seed": 7, "max_R_eigenvalue": 30.227238142474327, "predicted_mean_shrink": 1.430343664080256, "observed_mean_shrink": 0.45656432355707843 } ] }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "result": { "mean": 6.776016294956207, "std": 0.3423025857087886, "per_seed": [ 6.252743244171143, 7.0802903175354, 6.731771469116211, 6.9929890632629395, 6.439003944396973, 6.6504974365234375, 6.667675018310547, 7.393159866333008 ], "n": 8 }, "signature_records": [ { "seed": 0, "max_R_eigenvalue": 19.40545124327589, "predicted_mean_shrink": 1.5662079346101938, "observed_mean_shrink": 0.5596839399958696 }, { "seed": 1, "max_R_eigenvalue": 18.18841998175957, "predicted_mean_shrink": 1.5452502825983927, "observed_mean_shrink": 0.5358931141930726 }, { "seed": 2, "max_R_eigenvalue": 18.48060331058384, "predicted_mean_shrink": 1.596801548354272, "observed_mean_shrink": 0.5895995070263204 }, { "seed": 3, "max_R_eigenvalue": 18.32546811217679, "predicted_mean_shrink": 1.5674429841867479, "observed_mean_shrink": 0.5633915474229879 }, { "seed": 4, "max_R_eigenvalue": 22.240504809382237, "predicted_mean_shrink": 1.6088249246045303, "observed_mean_shrink": 0.6079893217536149 }, { "seed": 5, "max_R_eigenvalue": 15.727164040816485, "predicted_mean_shrink": 1.5816451385772001, "observed_mean_shrink": 0.5776791584829287 }, { "seed": 6, "max_R_eigenvalue": 21.040818460169987, "predicted_mean_shrink": 1.5729334873273007, "observed_mean_shrink": 0.5678789677304293 }, { "seed": 7, "max_R_eigenvalue": 18.48838750187182, "predicted_mean_shrink": 1.567401702259996, "observed_mean_shrink": 0.5609064495439936 } ] }, { "cfg": { "lr": 0.006, "weight_decay": 0.0001 }, "result": { "mean": 6.396751523017883, "std": 0.5465600136268931, "per_seed": [ 5.679222583770752, 6.877334117889404, 5.647143363952637, 6.208988666534424, 6.021716594696045, 6.866507053375244, 6.70987606048584, 7.163223743438721 ], "n": 8 }, "signature_records": [ { "seed": 0, "max_R_eigenvalue": 21.520722249039736, "predicted_mean_shrink": 1.5998354212422041, "observed_mean_shrink": 0.5803018674852735 }, { "seed": 1, "max_R_eigenvalue": 26.758464379751487, "predicted_mean_shrink": 1.568503006010591, "observed_mean_shrink": 0.5511755243066208 }, { "seed": 2, "max_R_eigenvalue": 21.52823242514386, "predicted_mean_shrink": 1.6324191170696345, "observed_mean_shrink": 0.6153089706860777 }, { "seed": 3, "max_R_eigenvalue": 19.959858225483167, "predicted_mean_shrink": 1.5989631426785875, "observed_mean_shrink": 0.5828625680472822 }, { "seed": 4, "max_R_eigenvalue": 21.21196735513266, "predicted_mean_shrink": 1.6179572148072678, "observed_mean_shrink": 0.6018178223934545 }, { "seed": 5, "max_R_eigenvalue": 21.802852810772954, "predicted_mean_shrink": 1.609761440973835, "observed_mean_shrink": 0.595776442199383 }, { "seed": 6, "max_R_eigenvalue": 22.262030674047317, "predicted_mean_shrink": 1.5987779728372298, "observed_mean_shrink": 0.5819983240302995 }, { "seed": 7, "max_R_eigenvalue": 22.797705272860483, "predicted_mean_shrink": 1.6093046685842605, "observed_mean_shrink": 0.594072235302656 } ] } ], "math_verification": { "max_abs_hessian_error": 1.0704046315979099e-08, "min_R_eigenvalue": 0.052625321728098556, "identity_pass": true }, "device": "cuda", "protocol_note": "Baseline sweep uses all idea learning rates and both Adam weight decays; 8 paired seeds.", "mechanism_signature": { "quantity": "Schur trust shrinkage measured during trained MLP updates", "predicted_vs_observed": { "predicted_mean_shrink": 1.6044402480254512, "observed_mean_shrink": 0.5879142193063809, "relative_error": 1.7290380047562637 }, "max_R_eigenvalue_mean": 22.230229174028956, "confirmed": false } }