1{
  2  "bench_version": 1,
  3  "track": "dynamics",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "epochs": 12
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 12
 17        },
 18        "mean": 0.004298779065720737
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 12
 24        },
 25        "mean": 0.0034615940821822733
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "epochs": 12
 31        },
 32        "mean": 0.0014865802077110857
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.0014198302378645167,
 37      "std": 0.00039839114901047844,
 38      "per_seed": [
 39        0.0009327387670055032,
 40        0.0018641706556081772,
 41        0.0010785290505737066,
 42        0.0020708823576569557,
 43        0.0009260809747502208,
 44        0.0014055331703275442,
 45        0.0016443432541564107,
 46        0.001436363672837615
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.0006506168301712023,
 53    "std": 0.00039993469646715593,
 54    "per_seed": [
 55      0.00010142059181816876,
 56      0.0005146179464645684,
 57      0.001238752854987979,
 58      0.001161463325843215,
 59      0.0002165312325814739,
 60      0.0007791328243911266,
 61      0.0008627780480310321,
 62      0.0003302378172520548
 63    ],
 64    "n": 8,
 65    "selected_cfg": {
 66      "lr": 0.01,
 67      "epochs": 12,
 68      "scale": 0.5
 69    }
 70  },
 71  "comparison": {
 72    "delta_mean": -0.0007692134076933144,
 73    "idea_wins": 7,
 74    "n_pairs": 8,
 75    "per_seed_diffs": [
 76      -0.0008313181751873344,
 77      -0.0013495527091436088,
 78      0.0001602238044142723,
 79      -0.0009094190318137407,
 80      -0.0007095497421687469,
 81      -0.0006264003459364176,
 82      -0.0007815652061253786,
 83      -0.0011061258555855602
 84    ],
 85    "p_value": 0.0163,
 86    "mde": 0.000368799706207264,
 87    "mde_rel_pct": 25.974915618218837,
 88    "verdict": "idea better (significant)",
 89    "system_worked": true
 90  },
 91  "math_check": {
 92    "local_bijective": true,
 93    "local_conservation_errors": [
 94      0,
 95      0,
 96      0,
 97      0,
 98      0,
 99      0
100    ],
101    "max_abs_conservation_error": 0
102  },
103  "idea_sweep": [
104    {
105      "cfg": {
106        "lr": 0.01,
107        "epochs": 12,
108        "scale": 0.5
109      },
110      "result": {
111        "mean": 0.0006506168301712023,
112        "std": 0.00039993469646715593,
113        "per_seed": [
114          0.00010142059181816876,
115          0.0005146179464645684,
116          0.001238752854987979,
117          0.001161463325843215,
118          0.0002165312325814739,
119          0.0007791328243911266,
120          0.0008627780480310321,
121          0.0003302378172520548
122        ],
123        "n": 8,
124        "selected_cfg": {
125          "lr": 0.01,
126          "epochs": 12,
127          "scale": 0.5
128        }
129      }
130    },
131    {
132      "cfg": {
133        "lr": 0.01,
134        "epochs": 12,
135        "scale": 1.0
136      },
137      "result": {
138        "mean": 0.0011277017038082704,
139        "std": 0.0005102781821450115,
140        "per_seed": [
141          0.0007301242440007627,
142          0.001474376767873764,
143          0.0014355103485286236,
144          0.0018912762170657516,
145          0.0006100746686570346,
146          0.0010276326211169362,
147          0.0015459696296602488,
148          0.0003066491335630417
149        ],
150        "n": 8
151      }
152    },
153    {
154      "cfg": {
155        "lr": 0.01,
156        "epochs": 12,
157        "scale": 2.0
158      },
159      "result": {
160        "mean": 0.0017072468253900297,
161        "std": 0.0008234145200694133,
162        "per_seed": [
163          0.0011916327057406306,
164          0.002574339509010315,
165          0.0014427917776629329,
166          0.0016165744746103883,
167          0.0012496613198891282,
168          0.002416426781564951,
169          0.0029221479780972004,
170          0.0002444000565446913
171        ],
172        "n": 8
173      }
174    }
175  ],
176  "structural_match": "controlled damped pendulum rollout tests stability/control of sequential state evolution",
177  "mechanism_signature": {
178    "quantity": "symbol_weight + carrier_weight conservation over trained model tokens",
179    "predicted": 0.0,
180    "observed": 0.0,
181    "absolute_error": 0.0,
182    "confirmed": true
183  }
184}