{
  "bench_version": 1,
  "track": "sequence",
  "model": "modular_sequence_local",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.003,
      "epochs": 12,
      "batch": 128
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 12,
          "batch": 128
        },
        "mean": 0.15198808163404465
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 12,
          "batch": 128
        },
        "mean": 0.1417534351348877
      },
      {
        "cfg": {
          "lr": 0.01,
          "epochs": 12,
          "batch": 128
        },
        "mean": 0.15112163126468658
      }
    ],
    "full": {
      "mean": 0.13169990107417107,
      "std": 0.017087285342305542,
      "per_seed": [
        0.1741165667772293,
        0.12635251879692078,
        0.13050207495689392,
        0.13604258000850677,
        0.11734391003847122,
        0.11690718680620193,
        0.1270616501569748,
        0.1252727210521698
      ],
      "n": 8
    }
  },
  "idea": {
    "per_seed": [
      0.1741165667772293,
      0.12635251879692078,
      0.13050207495689392,
      0.13604258000850677,
      0.11734391003847122,
      0.11690718680620193,
      0.1270616501569748,
      0.1252727210521698
    ],
    "mean": 0.13169990107417107,
    "std": 0.017087285342305542,
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.0,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0,
      0.0
    ],
    "p_value": 1.0,
    "mde": 0.0,
    "mde_rel_pct": 0.0,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "track_justification": "sequence is structurally matched: multi-token window correlations and a sequence-level model",
  "idea_sweep": [
    {
      "cfg": {
        "lr": 0.001,
        "epochs": 12,
        "batch": 128
      },
      "mean": 0.14950868487358093
    },
    {
      "cfg": {
        "lr": 0.003,
        "epochs": 12,
        "batch": 128
      },
      "mean": 0.13169990107417107
    },
    {
      "cfg": {
        "lr": 0.01,
        "epochs": 12,
        "batch": 128
      },
      "mean": 0.14627121202647686
    }
  ],
  "canonical_proposal": [
    "g3",
    "g1",
    "global",
    "g0",
    "g2"
  ],
  "canonical_word": [
    "g1",
    "g3",
    "global",
    "g0",
    "g2"
  ],
  "mechanism_signature": {
    "prediction": "disjoint-support swaps preserve the trained network output; group/global swaps do not",
    "predicted_commuting_max": 1e-05,
    "observed_commuting_max": 0.0,
    "predicted_noncommuting_positive": true,
    "observed_noncommuting_median": 1.8506829738616943,
    "confirmed": true,
    "n_trained_models": 8
  }
}