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    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 0.001552814821479842
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0009541053732391447
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.01
 27        },
 28        "mean": 0.0004885061644017696
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.00041624879668233916,
 33      "std": 0.00018933904234880896,
 34      "per_seed": [
 35        0.0003217894700355828,
 36        0.00039041441050358117,
 37        0.0003688229189720005,
 38        0.0008729978580959141,
 39        0.00046312736230902374,
 40        0.00017730507533997297,
 41        0.00040199048817157745,
 42        0.0003335427900310606
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.6313471160829067,
 49    "std": 0.09825334037840372,
 50    "per_seed": [
 51      0.5460999608039856,
 52      0.6500953435897827,
 53      0.7589279413223267,
 54      0.6426548957824707,
 55      0.5884038805961609,
 56      0.44923338294029236,
 57      0.7648789286613464,
 58      0.6504825949668884
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": 0.6309308672862244,
 64    "idea_wins": 0,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      0.54577817133395,
 68      0.6497049291792791,
 69      0.7585591184033547,
 70      0.6417818979243748,
 71      0.5879407532338519,
 72      0.4490560778649524,
 73      0.7644769381731749,
 74      0.6501490521768574
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.08777660086106652,
 78    "mde_rel_pct": 21087.532639296336,
 79    "verdict": "idea worse (significant)",
 80    "system_worked": false
 81  },
 82  "idea_sweep": [
 83    {
 84      "cfg": {
 85        "lr": 0.001
 86      },
 87      "result": {
 88        "mean": 0.8622342199087143,
 89        "std": 0.23361338646282398,
 90        "per_seed": [
 91          1.0948189496994019,
 92          1.3147796392440796,
 93          0.8199796676635742,
 94          0.7778266072273254,
 95          0.628410816192627,
 96          0.5279790759086609,
 97          0.9013363122940063,
 98          0.8327426910400391
 99        ],
100        "n": 8
101      }
102    },
103    {
104      "cfg": {
105        "lr": 0.003
106      },
107      "result": {
108        "mean": 0.6572596728801727,
109        "std": 0.11384008699978249,
110        "per_seed": [
111          0.7612924575805664,
112          0.5553685426712036,
113          0.712843120098114,
114          0.794246256351471,
115          0.6013367772102356,
116          0.4265899658203125,
117          0.6977173686027527,
118          0.7086828947067261
119        ],
120        "n": 8
121      }
122    },
123    {
124      "cfg": {
125        "lr": 0.01
126      },
127      "result": {
128        "mean": 0.6313471160829067,
129        "std": 0.09825334037840372,
130        "per_seed": [
131          0.5460999608039856,
132          0.6500953435897827,
133          0.7589279413223267,
134          0.6426548957824707,
135          0.5884038805961609,
136          0.44923338294029236,
137          0.7648789286613464,
138          0.6504825949668884
139        ],
140        "n": 8
141      }
142    }
143  ],
144  "selection_note": "same lr union and epochs/batch for both systems",
145  "mechanism_signature": {
146    "prediction": "trained fold transition remains approximately reversible away from singularity",
147    "predicted_reconstruction_error": "near numerical precision (ideal fixed fold)",
148    "observed_reconstruction_error": 5.6109079310529e-07,
149    "observed_state_scale": 1.3907469511032104,
150    "observed_reverse_forward_norm_ratio": 1.000000074505806,
151    "confirmed": true
152  },
153  "custom_track": null
154}