{ "bench_version": 1, "track": "tabular", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "weight_decay": 0.001 }, "sweep": [ { "cfg": { "lr": 0.0015, "weight_decay": 0.0 }, "mean": 14.94258713722229 }, { "cfg": { "lr": 0.0015, "weight_decay": 0.0001 }, "mean": 14.863693475723267 }, { "cfg": { "lr": 0.0015, "weight_decay": 0.001 }, "mean": 14.837073802947998 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 11.12595009803772 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "mean": 10.99112582206726 }, { "cfg": { "lr": 0.003, "weight_decay": 0.001 }, "mean": 10.882298469543457 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0 }, "mean": 7.924989342689514 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0001 }, "mean": 7.200156211853027 }, { "cfg": { "lr": 0.006, "weight_decay": 0.001 }, "mean": 7.055913209915161 } ], "full": { "mean": 7.258670151233673, "std": 0.43432421459398, "per_seed": [ 6.749958515167236, 7.283852577209473, 7.195377826690674, 6.994463920593262, 6.717870235443115, 7.567739486694336, 7.43061637878418, 8.12948226928711 ], "n": 8 } }, "idea": { "mean": 12.942374348640442, "std": 0.7042793380398409, "per_seed": [ 13.581195831298828, 13.057331085205078, 13.15151309967041, 11.582756042480469, 12.118292808532715, 13.407008171081543, 12.833832740783691, 13.8070650100708 ], "n": 8 }, "comparison": { "delta_mean": 5.683704197406769, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 6.831237316131592, 5.7734785079956055, 5.956135272979736, 4.588292121887207, 5.4004225730896, 5.839268684387207, 5.403216361999512, 5.677582740783691 ], "p_value": 0.0081, "mde": 0.5271462525241571, "mde_rel_pct": 7.262297935312079, "verdict": "idea worse (significant)", "system_worked": false }, "idea_trials": [ { "cfg": { "lr": 0.003, "weight_decay": 0.0001, "target_sigma": 2.0 }, "result": { "mean": 16.349509477615356, "std": 0.9009887062653857, "per_seed": [ 16.687042236328125, 16.01118278503418, 16.832225799560547, 14.972037315368652, 15.328200340270996, 16.66720962524414, 16.223684310913086, 18.074493408203125 ], "n": 8 } }, { "cfg": { "lr": 0.0015, "weight_decay": 0.0001, "target_sigma": 2.0 }, "result": { "mean": 27.306380033493042, "std": 3.1792541193528128, "per_seed": [ 27.235408782958984, 30.38294792175293, 23.430057525634766, 23.617746353149414, 27.093000411987305, 24.0976505279541, 30.76529884338379, 31.828929901123047 ], "n": 8 } }, { "cfg": { "lr": 0.006, "weight_decay": 0.0001, "target_sigma": 2.0 }, "result": { "mean": 12.942374348640442, "std": 0.7042793380398409, "per_seed": [ 13.581195831298828, 13.057331085205078, 13.15151309967041, 11.582756042480469, 12.118292808532715, 13.407008171081543, 12.833832740783691, 13.8070650100708 ], "n": 8 } } ], "protocol_note": "tabular is structurally matched: regularization/stability of linear layers.", "mechanism_signature": { "per_seed": [ { "predicted_bound_sigma": [ 1.9660207790324025, 1.9599940777246498, 1.7198881439758398 ], "observed_true_sigma": [ 1.965899109840393, 1.9599742889404297, 1.7198880910873413 ], "max_bound_minus_observed": 0.0001216691920093993, "target_sigma": 2.0, "confirmed": true }, { "predicted_bound_sigma": [ 1.987042231331432, 1.9654762301286028, 1.8110116015778066 ], "observed_true_sigma": [ 1.9857813119888306, 1.965454339981079, 1.8110116720199585 ], "max_bound_minus_observed": 0.0012609193426014187, "target_sigma": 2.0, "confirmed": true }, { "predicted_bound_sigma": [ 1.9618061463008163, 1.9643066088926524, 1.7638919009485563 ], "observed_true_sigma": [ 1.9616186618804932, 1.9643056392669678, 1.7638919353485107 ], "max_bound_minus_observed": 0.00018748442032312518, "target_sigma": 2.0, "confirmed": true }, { "predicted_bound_sigma": [ 2.0027832775795726, 1.9854929319578933, 1.7735200867709888 ], "observed_true_sigma": [ 2.00002384185791, 1.9853485822677612, 1.7735198736190796 ], "max_bound_minus_observed": 0.0027594357216624843, "target_sigma": 2.0, "confirmed": true }, { "predicted_bound_sigma": [ 1.9771301980155285, 1.9942939741761676, 1.7462854647292247 ], "observed_true_sigma": [ 1.9769728183746338, 1.9943008422851562, 1.7462854385375977 ], "max_bound_minus_observed": 0.0001573796408946837, "target_sigma": 2.0, "confirmed": true }, { "predicted_bound_sigma": [ 2.034734505621217, 2.0299429975184977, 1.743557330086019 ], "observed_true_sigma": [ 2.034123420715332, 2.0295772552490234, 1.743557095527649 ], "max_bound_minus_observed": 0.0006110849058851286, "target_sigma": 2.0, "confirmed": true }, { "predicted_bound_sigma": [ 1.9666434720553738, 1.9534572174409695, 1.7779871345230767 ], "observed_true_sigma": [ 1.9653239250183105, 1.9534378051757812, 1.7779872417449951 ], "max_bound_minus_observed": 0.0013195470370632734, "target_sigma": 2.0, "confirmed": true }, { "predicted_bound_sigma": [ 2.0642188408904656, 2.0205151771620473, 1.6440712816243759 ], "observed_true_sigma": [ 2.064063787460327, 2.020517587661743, 1.6440712213516235 ], "max_bound_minus_observed": 0.00015505343013844097, "target_sigma": 2.0, "confirmed": true } ], "predicted_bound_sigma_mean": 1.9089196504193406, "observed_true_sigma_mean": 1.9086223244667053, "max_bound_minus_observed": 0.0027594357216624843, "confirmed": true }, "math_check": { "max_upper_bound_violation": 0.0, "mean_bound_gap_squared": 1.9329186102939733, "clustered_recovery_abs_error": 0.0, "passed": true } }