{ "bench_report": { "baseline": { "best_cfg": { "lr": 0.003 }, "full": { "mean": 0.4984375163912773, "n": 8, "per_seed": [ 0.48750001192092896, 0.5062500238418579, 0.48125001788139343, 0.4937500059604645, 0.5562500357627869, 0.4625000059604645, 0.5062500238418579, 0.4937500059604645 ], "std": 0.025531468853458108 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.5187500193715096 }, { "cfg": { "lr": 0.003 }, "mean": 0.4921875149011612 }, { "cfg": { "lr": 0.01 }, "mean": 0.4984375163912773 } ] }, "bench_version": 1, "comparison": { "delta_mean": -0.4984375163912773, "idea_wins": 8, "mde": 0.0228222220228749, "mde_rel_pct": 4.578752857150359, "n_pairs": 8, "p_value": 0.0081, "per_seed_diffs": [ -0.48750001192092896, -0.5062500238418579, -0.48125001788139343, -0.4937500059604645, -0.5562500357627869, -0.4625000059604645, -0.5062500238418579, -0.4937500059604645 ], "system_worked": true, "verdict": "idea better (significant)" }, "idea": { "best_cfg": { "lr": 0.001 }, "mean": 0.0, "n": 8, "per_seed": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "std": 0.0, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.0 }, { "cfg": { "lr": 0.003 }, "mean": 0.0 }, { "cfg": { "lr": 0.01 }, "mean": 0.0 } ] }, "mechanism_signature": { "baseline_max_output_change": 0.4521815776824951, "baseline_mean_output_change": 0.02940310724079609, "confirmed": true, "idea_max_output_change": 1.7881393432617188e-07, "idea_mean_output_change": 3.8091179099808414e-09, "observed_feature_change": 0.0, "predicted_invariant_output_change": 0.0, "prediction": "P2 local-correlation inputs and their trained predictions should be unchanged by global rotation and independent apolar flips." }, "metric_direction": "lower is better", "model": "mlp_tiny", "n_seeds": 8, "track": "orientation_phase_quotient" }, "custom_track": { "domain": "molecular_orientation_symmetry", "file": "orientation_track.py", "name": "orientation_phase_quotient" }, "math_check": { "max_transformed_feature_error": 0.0, "predicted": 0.0 }, "protocol": { "batch": 128, "epochs": 18, "lr_union": [ 0.001, 0.003, 0.01 ], "paired_seeds": [ 0, 1, 2, 3, 4, 5, 6, 7 ] } }