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": 0.006
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.0015
 15        },
 16        "mean": 0.00198724222718738
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.003
 21        },
 22        "mean": 0.0013494549202732742
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.006
 27        },
 28        "mean": 0.0007156882056733593
 29      }
 30    ],
 31    "full": {
 32      "mean": 0.0006826339049439412,
 33      "std": 0.0002001734108210113,
 34      "per_seed": [
 35        0.0008196589769795537,
 36        0.0005301681812852621,
 37        0.000536333944182843,
 38        0.0009765917202457786,
 39        0.0009811640484258533,
 40        0.0005039663519710302,
 41        0.0006512094405479729,
 42        0.00046197857591323555
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 0.0005652364307024982,
 49    "std": 0.00019911234856491642,
 50    "per_seed": [
 51      0.0006373166106641293,
 52      0.0004831327241845429,
 53      0.0003901786112692207,
 54      0.0008194606634788215,
 55      0.0009116221335716546,
 56      0.0003751792246475816,
 57      0.0005717286258004606,
 58      0.00033327285200357437
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -0.00011739747424144298,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -0.00018234236631542444,
 68      -4.7035457100719213e-05,
 69      -0.00014615533291362226,
 70      -0.00015713105676695704,
 71      -6.95419148541987e-05,
 72      -0.00012878712732344866,
 73      -7.948081474751234e-05,
 74      -0.00012870572390966117
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 3.942211076832542e-05,
 78    "mde_rel_pct": 5.775000403995876,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "mechanism_signature": {
 83    "mechanism_signature": {
 84      "prediction": "encoder-reset should suppress state sensitivity; final perturbation ratio < 1",
 85      "observed": [
 86        {
 87          "mode": "carry",
 88          "test_mse": 0.001019460498355329,
 89          "observed_context_to_final_state_ratio": 0.008399932645261288
 90        },
 91        {
 92          "mode": "reset",
 93          "test_mse": 0.0011074204230681062,
 94          "observed_context_to_final_state_ratio": 0.008114841766655445
 95        }
 96      ],
 97      "confirmed": true
 98    },
 99    "idea_sweep": {
100      "0.0015": {
101        "mean": 0.001871054424555041,
102        "std": 0.0004904266878392466,
103        "per_seed": [
104          0.0014404993271455169,
105          0.001380899571813643,
106          0.0018803070997819304,
107          0.0025381247978657484,
108          0.0022744236048310995,
109          0.001058555906638503,
110          0.0021054912358522415,
111          0.0022901338525116444
112        ],
113        "n": 8
114      },
115      "0.003": {
116        "mean": 0.0011377038972568698,
117        "std": 0.00018237536205938462,
118        "per_seed": [
119          0.0011074204230681062,
120          0.0010416923323646188,
121          0.0010215630754828453,
122          0.0014193312963470817,
123          0.0012926014605909586,
124          0.0008008744916878641,
125          0.001297146431170404,
126          0.00112100166734308
127        ],
128        "n": 8
129      },
130      "0.006": {
131        "mean": 0.0005652364307024982,
132        "std": 0.00019911234856491642,
133        "per_seed": [
134          0.0006373166106641293,
135          0.0004831327241845429,
136          0.0003901786112692207,
137          0.0008194606634788215,
138          0.0009116221335716546,
139          0.0003751792246475816,
140          0.0005717286258004606,
141          0.00033327285200357437
142        ],
143        "n": 8
144      }
145    },
146    "chosen_lr": 0.006
147  },
148  "protocol_notes": "8 paired seeds; baseline sweep and idea sweep share lr union; n_train=400, n_test=160, 18 epochs, batch 128."
149}