{ "bench_version": 1, "track": "custom_correlated_multitask_regression", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.5403974950313568 }, { "cfg": { "lr": 0.003 }, "mean": 0.39297179132699966 }, { "cfg": { "lr": 0.006 }, "mean": 0.3831307291984558 } ], "full": { "mean": 0.3909529335796833, "std": 0.020058499256199364, "per_seed": [ 0.3602304756641388, 0.3643752634525299, 0.40289655327796936, 0.4050206243991852, 0.420786589384079, 0.38486582040786743, 0.4063948094844818, 0.38305333256721497 ], "n": 8 } }, "idea": { "mean": 0.3923187702894211, "std": 0.02027321236177226, "per_seed": [ 0.3610771894454956, 0.36850693821907043, 0.40293797850608826, 0.4014033079147339, 0.42628708481788635, 0.38414520025253296, 0.4089997410774231, 0.38519272208213806 ], "n": 8 }, "comparison": { "delta_mean": 0.0013658367097377777, "idea_wins": 2, "n_pairs": 8, "per_seed_diffs": [ 0.0008467137813568115, 0.004131674766540527, 4.1425228118896484e-05, -0.003617316484451294, 0.005500495433807373, -0.0007206201553344727, 0.002604931592941284, 0.0021393895149230957 ], "p_value": 0.22765, "mde": 0.002408680129708675, "mde_rel_pct": 0.6161048870139102, "verdict": "no measurable effect", "system_worked": false }, "mechanism_signature": { "prediction": "residual-TC clustering discovers correlated task pairs and lowers within-cluster residual dependence", "predicted_mean_all_tc": 1.3089573328100736, "observed_mean_within_cluster_tc": 0.403780898657931, "pair_recovery_rate": 1.0, "confirmed": true }, "custom_track": { "name": "correlated_multitask_regression", "file": "custom_multitask_track.py", "domain": "multi_task_learning" }, "idea_sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.5575724691152573 }, { "cfg": { "lr": 0.003 }, "mean": 0.4013301022350788 }, { "cfg": { "lr": 0.006 }, "mean": 0.3923187702894211 } ] }