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": 0.001
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.4451246112585068
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.4927415996789932
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 1.6619523614645004
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.45439932495355606,
 33      "std": 0.04081052029431072,
 34      "per_seed": [
 35        0.4278562068939209,
 36        0.3956301808357239,
 37        0.46248915791511536,
 38        0.49452289938926697,
 39        0.44014036655426025,
 40        0.4744090139865875,
 41        0.5266208648681641,
 42        0.41352590918540955
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 1.0940600484609604,
 49    "std": 0.0844539494160516,
 50    "per_seed": [
 51      1.086983323097229,
 52      0.9213972091674805,
 53      1.21669602394104,
 54      1.1440339088439941,
 55      1.0787184238433838,
 56      1.0460021495819092,
 57      1.177429437637329,
 58      1.0812199115753174
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.6396607235074043,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.6591271162033081,
 68      0.5257670283317566,
 69      0.7542068660259247,
 70      0.6495110094547272,
 71      0.6385780572891235,
 72      0.5715931355953217,
 73      0.650808572769165,
 74      0.6676940023899078
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.056634104586998066,
 78    "mde_rel_pct": 12.463509841874568,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "mechanism_signature": {
 83    "track_match": "multi-token sequence forecast",
 84    "mean_events_per_sample": 1536.0,
 85    "mean_shared_node_visits": 32.0,
 86    "predicted_shared_visits": 32.0,
 87    "observed_to_predicted_ratio": 1.0,
 88    "source_message_abs_mean": 0.6570842266082764,
 89    "prediction": "each sequence token visits every shared hidden node once",
 90    "confirmed": true
 91  },
 92  "idea_sweep": [
 93    {
 94      "cfg": {
 95        "lr": 0.001
 96      },
 97      "mean": 1.0940600484609604
 98    },
 99    {
100      "cfg": {
101        "lr": 0.003
102      },
103      "mean": 1.0993548929691315
104    },
105    {
106      "cfg": {
107        "lr": 0.01
108      },
109      "mean": 1.121301993727684
110    }
111  ],
112  "protocol_notes": {
113    "epochs": 12,
114    "n_train": 400,
115    "n_test": 200,
116    "baseline_grid": [
117      {
118        "lr": 0.001
119      },
120      {
121        "lr": 0.003
122      },
123      {
124        "lr": 0.01
125      }
126    ],
127    "idea_grid": [
128      {
129        "lr": 0.001
130      },
131      {
132        "lr": 0.003
133      },
134      {
135        "lr": 0.01
136      }
137    ],
138    "architecture": "transformer baseline vs trainable delayed shared-accumulator graph"
139  }
140}