1{
  2  "bench_version": 1,
  3  "track": "relational_dynamics",
  4  "model": "pair_temporal_mlp",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.8927807062864304
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.8921257555484772
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.8923652321100235
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.9345984682440758,
 33      "std": 0.08055134979477402,
 34      "per_seed": [
 35        0.946091890335083,
 36        0.782263457775116,
 37        0.9252486824989319,
 38        0.9148989915847778,
 39        1.0442157983779907,
 40        1.0487204790115356,
 41        0.8793458342552185,
 42        0.9360026121139526
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.014908512006513774,
 49    "std": 0.0029079161802027706,
 50    "per_seed": [
 51      0.012118596583604813,
 52      0.013483788818120956,
 53      0.013024883344769478,
 54      0.01563272811472416,
 55      0.011209259741008282,
 56      0.017514415085315704,
 57      0.015654947608709335,
 58      0.020629476755857468
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.919689956237562,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.9339732937514782,
 68      -0.768779668956995,
 69      -0.9122237991541624,
 70      -0.8992662634700537,
 71      -1.0330065386369824,
 72      -1.03120606392622,
 73      -0.8636908866465092,
 74      -0.9153731353580952
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.0719854515760221,
 78    "mde_rel_pct": 7.702286492216108,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "custom_track": {
 83    "name": "relational_dynamics",
 84    "file": "relational_dynamics_track.py",
 85    "domain": "dynamics"
 86  },
 87  "idea_config": {
 88    "lr": 0.01
 89  },
 90  "idea_sweep": [
 91    {
 92      "cfg": {
 93        "lr": 0.001
 94      },
 95      "result": {
 96        "mean": 0.6230705045163631,
 97        "std": 0.134810917972084,
 98        "per_seed": [
 99          0.7404102087020874,
100          0.42426830530166626,
101          0.6833330392837524,
102          0.4215412437915802,
103          0.5891903042793274,
104          0.7319473028182983,
105          0.5875273942947388,
106          0.8063462376594543
107        ],
108        "n": 8
109      }
110    },
111    {
112      "cfg": {
113        "lr": 0.003
114      },
115      "result": {
116        "mean": 0.09541938733309507,
117        "std": 0.014735930247991384,
118        "per_seed": [
119          0.10757716745138168,
120          0.07397669553756714,
121          0.07694181054830551,
122          0.0862903743982315,
123          0.09326057136058807,
124          0.11803340911865234,
125          0.10881876945495605,
126          0.09845630079507828
127        ],
128        "n": 8
129      }
130    },
131    {
132      "cfg": {
133        "lr": 0.01
134      },
135      "result": {
136        "mean": 0.014908512006513774,
137        "std": 0.0029079161802027706,
138        "per_seed": [
139          0.012118596583604813,
140          0.013483788818120956,
141          0.013024883344769478,
142          0.01563272811472416,
143          0.011209259741008282,
144          0.017514415085315704,
145          0.015654947608709335,
146          0.020629476755857468
147        ],
148        "n": 8
149      }
150    }
151  ],
152  "math_check": {
153    "inactive_retention_max_abs_error": 0.0,
154    "gru_outputs_finite": true,
155    "retention_exact": true
156  },
157  "mechanism_signature": {
158    "prediction": "persistent pair state should reduce reactivation error after early evidence is masked",
159    "baseline_reactivation_mse": 0.9345984682440758,
160    "idea_reactivation_mse": 0.3951987475156784,
161    "observed_ratio_idea_over_baseline": 0.4228540501015137,
162    "confirmed": true
163  }
164}