1{
  2  "bench_version": 1,
  3  "track": "sequence",
  4  "model": "transformer_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.006,
 10      "eps": 0.7,
 11      "iters": 12
 12    },
 13    "sweep": [
 14      {
 15        "cfg": {
 16          "lr": 0.001,
 17          "eps": 0.7,
 18          "iters": 12
 19        },
 20        "mean": 0.9074711948633194
 21      },
 22      {
 23        "cfg": {
 24          "lr": 0.003,
 25          "eps": 0.7,
 26          "iters": 12
 27        },
 28        "mean": 0.915537878870964
 29      },
 30      {
 31        "cfg": {
 32          "lr": 0.006,
 33          "eps": 0.7,
 34          "iters": 12
 35        },
 36        "mean": 0.8664272725582123
 37      }
 38    ],
 39    "full": {
 40      "mean": 0.9269421473145485,
 41      "std": 0.08130362337327321,
 42      "per_seed": [
 43        0.9268826246261597,
 44        0.8607907891273499,
 45        0.8572450280189514,
 46        0.8207906484603882,
 47        1.0688729286193848,
 48        0.9266430139541626,
 49        1.0371272563934326,
 50        0.9171848893165588
 51      ],
 52      "n": 8
 53    }
 54  },
 55  "idea": {
 56    "mean": 0.9356912449002266,
 57    "std": 0.05696883264914271,
 58    "per_seed": [
 59      0.9681993126869202,
 60      0.8495118021965027,
 61      0.900809645652771,
 62      0.961909830570221,
 63      0.9716687798500061,
 64      0.9182450175285339,
 65      1.038270115852356,
 66      0.876915454864502
 67    ],
 68    "n": 8
 69  },
 70  "comparison": {
 71    "delta_mean": 0.0087490975856781,
 72    "idea_wins": 4,
 73    "n_pairs": 8,
 74    "per_seed_diffs": [
 75      0.0413166880607605,
 76      -0.011278986930847168,
 77      0.04356461763381958,
 78      0.14111918210983276,
 79      -0.09720414876937866,
 80      -0.008397996425628662,
 81      0.0011428594589233398,
 82      -0.040269434452056885
 83    ],
 84    "p_value": 0.75565,
 85    "mde": 0.05842531038665775,
 86    "mde_rel_pct": 6.303015841487215,
 87    "verdict": "no measurable effect",
 88    "system_worked": false
 89  },
 90  "idea_sweep": [
 91    {
 92      "lr": 0.006,
 93      "eps": 0.5,
 94      "iters": 12
 95    },
 96    {
 97      "lr": 0.006,
 98      "eps": 0.7,
 99      "iters": 12
100    },
101    {
102      "lr": 0.006,
103      "eps": 1.0,
104      "iters": 12
105    }
106  ],
107  "track_justification": "Sequence forecast contains multi-token correlations and directly exercises attention.",
108  "mechanism_signature": {
109    "predicted_max_marginal_error": 0.0001,
110    "observed_mean_row_marginal_error": 0.007857249001972377,
111    "observed_mean_column_marginal_error": 1.1294105761727735e-08,
112    "observed_min_attention_mean": 1.0011145091900235e-14,
113    "confirmed": false
114  }
115}