Transport-PDE Predictor for Delayed Neural State Updates / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  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": 0.009
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 3.668733984341088e-05
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 8.922007737055537e-06
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.009
 27        },
 28        "mean": 4.542415013020218e-06
 29      }
 30    ],
 31    "full": {
 32      "mean": 5.152216147052968e-06,
 33      "std": 1.9023458106411929e-06,
 34      "per_seed": [
 35        4.7863295549177565e-06,
 36        5.878216143173631e-06,
 37        3.588439312807168e-06,
 38        3.7736158446932677e-06,
 39        6.708174169034464e-06,
 40        8.705701475264505e-06,
 41        5.577248430199688e-06,
 42        2.2000042463332647e-06
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 4.707826093408585e-06,
 49    "std": 2.320462925135068e-06,
 50    "per_seed": [
 51      2.252499598398572e-06,
 52      3.3756230095605133e-06,
 53      3.7817651445948286e-06,
 54      8.986970897240099e-06,
 55      3.874472895404324e-06,
 56      3.393555971342721e-06,
 57      8.273613275378011e-06,
 58      3.7241079553496093e-06
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -4.4439005364438344e-07,
 64    "idea_wins": 4,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -2.5338299565191846e-06,
 68      -2.502593133613118e-06,
 69      1.9332583178766072e-07,
 70      5.213355052546831e-06,
 71      -2.8337012736301403e-06,
 72      -5.3121455039217835e-06,
 73      2.696364845178323e-06,
 74      1.5241037090163445e-06
 75    ],
 76    "p_value": 0.7344,
 77    "mde": 2.8993112919196206e-06,
 78    "mde_rel_pct": 56.27309121295322,
 79    "verdict": "no significant win",
 80    "system_worked": false
 81  },
 82  "idea_sweep": [
 83    {
 84      "cfg": {
 85        "lr": 0.001
 86      },
 87      "result": {
 88        "mean": 3.7107075286257896e-05,
 89        "std": 6.4375009711858975e-06,
 90        "per_seed": [
 91          2.8180171284475364e-05,
 92          4.052478107041679e-05,
 93          3.2498250220669433e-05,
 94          5.090365812066011e-05,
 95          3.462454697000794e-05,
 96          3.355262742843479e-05,
 97          3.641599687398411e-05,
 98          4.015657032141462e-05
 99        ],
100        "n": 8
101      }
102    },
103    {
104      "cfg": {
105        "lr": 0.003
106      },
107      "result": {
108        "mean": 1.0382680102338782e-05,
109        "std": 3.1424418893375147e-06,
110        "per_seed": [
111          1.1441469723649789e-05,
112          9.949631021299865e-06,
113          1.4452817595156375e-05,
114          1.5155372238950804e-05,
115          8.438014447165187e-06,
116          5.9799863265652675e-06,
117          1.1092530257883482e-05,
118          6.551619208039483e-06
119        ],
120        "n": 8
121      }
122    },
123    {
124      "cfg": {
125        "lr": 0.009
126      },
127      "result": {
128        "mean": 4.707826093408585e-06,
129        "std": 2.320462925135068e-06,
130        "per_seed": [
131          2.252499598398572e-06,
132          3.3756230095605133e-06,
133          3.7817651445948286e-06,
134          8.986970897240099e-06,
135          3.874472895404324e-06,
136          3.393555971342721e-06,
137          8.273613275378011e-06,
138          3.7241079553496093e-06
139        ],
140        "n": 8
141      }
142    }
143  ],
144  "track_justification": "Dynamics is the built-in control/stability track and contains delayed actuated pendulum rollouts.",
145  "mechanism_signature": {
146    "delay": 4,
147    "trained_models": true,
148    "measured_on": "official dynamics test windows",
149    "mean_predicted_terminal_state_shift": 0.21677356958389282,
150    "predicted_signature": "finite-horizon transport prediction changes stale terminal state",
151    "confirmed": true
152  }
153}