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": 0.004298779065720737
17      },
18      {
19        "cfg": {
20          "lr": 0.003
21        },
22        "mean": 0.0034615940821822733
23      },
24      {
25        "cfg": {
26          "lr": 0.01
27        },
28        "mean": 0.0014865802077110857
29      }
30    ],
31    "full": {
32      "mean": 0.0014198302378645167,
33      "std": 0.00039839114901047844,
34      "per_seed": [
35        0.0009327387670055032,
36        0.0018641706556081772,
37        0.0010785290505737066,
38        0.0020708823576569557,
39        0.0009260809747502208,
40        0.0014055331703275442,
41        0.0016443432541564107,
42        0.001436363672837615
43      ],
44      "n": 8
45    }
46  },
47  "idea": {
48    "mean": 0.02781403553672135,
49    "std": 0.005121785167552724,
50    "per_seed": [
51      0.02624191716313362,
52      0.019634274765849113,
53      0.030885715037584305,
54      0.03626050800085068,
55      0.020821113139390945,
56      0.029457034543156624,
57      0.030522819608449936,
58      0.028688902035355568
59    ],
60    "n": 8
61  },
62  "comparison": {
63    "delta_mean": 0.026394205298856832,
64    "idea_wins": 0,
65    "n_pairs": 8,
66    "per_seed_diffs": [
67      0.025309178396128118,
68      0.017770104110240936,
69      0.029807185987010598,
70      0.03418962564319372,
71      0.019895032164640725,
72      0.02805150137282908,
73      0.028878476354293525,
74      0.027252538362517953
75    ],
76    "p_value": 0.0081,
77    "mde": 0.004467906007621001,
78    "mde_rel_pct": 314.6788882550435,
79    "verdict": "idea worse (significant)",
80    "system_worked": false
81  },
82  "idea_cfg": {
83    "lr": 0.01
84  },
85  "protocol_note": "dynamics is structurally matched: controlled pendulum rollout",
86  "mechanism_signature": {
87    "prediction": "mobility-weighted action should scale residual energy by inverse positive mobility",
88    "predicted_inverse_mobility_mean": "not directly averaged in the original run; mobility mean=0.79597 implies inverse-mobility weighting > 1",
89    "observed_residual_mse": 5.364831447601318,
90    "observed_mobility_weighted_residual": 6.362071514129639,
91    "observed_weighted_to_unweighted_ratio": 1.1858846967082626,
92    "predicted_direction": "weighted residual exceeds unweighted residual because measured mobility is below 1",
93    "confirmed": true
94  }
95}