1{
  2  "bench_version": 1,
  3  "track": "custom_correlated_multitask_regression",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.5403974950313568
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.39297179132699966
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.3831307291984558
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.3909529335796833,
 33      "std": 0.020058499256199364,
 34      "per_seed": [
 35        0.3602304756641388,
 36        0.3643752634525299,
 37        0.40289655327796936,
 38        0.4050206243991852,
 39        0.420786589384079,
 40        0.38486582040786743,
 41        0.4063948094844818,
 42        0.38305333256721497
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.3923187702894211,
 49    "std": 0.02027321236177226,
 50    "per_seed": [
 51      0.3610771894454956,
 52      0.36850693821907043,
 53      0.40293797850608826,
 54      0.4014033079147339,
 55      0.42628708481788635,
 56      0.38414520025253296,
 57      0.4089997410774231,
 58      0.38519272208213806
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.0013658367097377777,
 64    "idea_wins": 2,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.0008467137813568115,
 68      0.004131674766540527,
 69      4.1425228118896484e-05,
 70      -0.003617316484451294,
 71      0.005500495433807373,
 72      -0.0007206201553344727,
 73      0.002604931592941284,
 74      0.0021393895149230957
 75    ],
 76    "p_value": 0.22765,
 77    "mde": 0.002408680129708675,
 78    "mde_rel_pct": 0.6161048870139102,
 79    "verdict": "no measurable effect",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "prediction": "residual-TC clustering discovers correlated task pairs and lowers within-cluster residual dependence",
 84    "predicted_mean_all_tc": 1.3089573328100736,
 85    "observed_mean_within_cluster_tc": 0.403780898657931,
 86    "pair_recovery_rate": 1.0,
 87    "confirmed": true
 88  },
 89  "custom_track": {
 90    "name": "correlated_multitask_regression",
 91    "file": "custom_multitask_track.py",
 92    "domain": "multi_task_learning"
 93  },
 94  "idea_sweep": [
 95    {
 96      "cfg": {
 97        "lr": 0.001
 98      },
 99      "mean": 0.5575724691152573
100    },
101    {
102      "cfg": {
103        "lr": 0.003
104      },
105      "mean": 0.4013301022350788
106    },
107    {
108      "cfg": {
109        "lr": 0.006
110      },
111      "mean": 0.3923187702894211
112    }
113  ]
114}