{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "epochs": 6,
      "alpha": 0.0,
      "weighted": false
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 6,
          "alpha": 0.0,
          "weighted": false
        },
        "mean": 0.15612634923309088
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 6,
          "alpha": 0.0,
          "weighted": false
        },
        "mean": 0.022599048796109855
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 6,
          "alpha": 0.0,
          "weighted": false
        },
        "mean": 0.013443514122627676
      }
    ],
    "full": {
      "mean": 0.013443514122627676,
      "std": 0.0034362747708833723,
      "per_seed": [
        0.015814311802387238,
        0.011003119871020317,
        0.014025969430804253,
        0.013154271990060806,
        0.01978541910648346,
        0.015274417586624622,
        0.010594896040856838,
        0.00789570715278387
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.016453448799438775,
    "std": 0.003337952311107532,
    "per_seed": [
      0.012871215119957924,
      0.01926228776574135,
      0.019986655563116074,
      0.01737268455326557,
      0.009503618814051151,
      0.017524683848023415,
      0.016284719109535217,
      0.018821725621819496
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.003009934676811099,
    "idea_wins": 2,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.0029430966824293137,
      0.008259167894721031,
      0.005960686132311821,
      0.004218412563204765,
      -0.010281800292432308,
      0.0022502662613987923,
      0.005689823068678379,
      0.010926018469035625
    ],
    "p_value": 0.24735,
    "mde": 0.005655673555602832,
    "mde_rel_pct": 42.06990452059994,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "mechanism_signature": {
      "metric": 0.012871215119957924,
      "critic_bce": 0.09358271956443787,
      "failure_rate": 0.0625,
      "predicted_enrichment": 2.64437198638916,
      "observed_enrichment": 2.55679988861084,
      "ess_fraction": 0.9292901097924411,
      "weighted_pool_loss": 0.013131698593497276,
      "uniform_pool_loss": 0.013131696730852127
    },
    "math_check": {
      "rows": [
        {
          "alpha": 0.5,
          "predicted_enrichment": 1.1935695059810845,
          "sampled_enrichment": 1.1971491228070177,
          "identity_l2": 2.180444593116775e-16,
          "ess_fraction": 0.7844566886686185
        },
        {
          "alpha": 1.0,
          "predicted_enrichment": 1.3319936587355161,
          "sampled_enrichment": 1.3263157894736841,
          "identity_l2": 1.6025315005387115e-16,
          "ess_fraction": 0.35981862086454786
        },
        {
          "alpha": 2.0,
          "predicted_enrichment": 1.5119809457893971,
          "sampled_enrichment": 1.5087719298245612,
          "identity_l2": 2.1315227208338676e-16,
          "ess_fraction": 0.08387276797729101
        }
      ],
      "max_identity_l2": 2.180444593116775e-16,
      "confirmed_identity": true
    },
    "prediction": "criticality proposal enrichment equals E_q[y]/E_p[y] while weighted expectation preserves uniform loss",
    "confirmed": true,
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 6,
          "alpha": 0.5,
          "weighted": true
        },
        "result": {
          "mean": 0.016453448799438775,
          "std": 0.003337952311107532,
          "per_seed": [
            0.012871215119957924,
            0.01926228776574135,
            0.019986655563116074,
            0.01737268455326557,
            0.009503618814051151,
            0.017524683848023415,
            0.016284719109535217,
            0.018821725621819496
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 6,
          "alpha": 1.0,
          "weighted": true
        },
        "result": {
          "mean": 0.01651376229710877,
          "std": 0.0028040435177098303,
          "per_seed": [
            0.01849322021007538,
            0.01740805059671402,
            0.017674783244729042,
            0.018753860145807266,
            0.014908691868185997,
            0.014080381020903587,
            0.010877873748540878,
            0.019913237541913986
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 6,
          "alpha": 2.0,
          "weighted": true
        },
        "result": {
          "mean": 0.01855568215250969,
          "std": 0.006122484793310474,
          "per_seed": [
            0.025685645639896393,
            0.020075352862477303,
            0.025499150156974792,
            0.0117742158472538,
            0.01591908372938633,
            0.008489308878779411,
            0.016218945384025574,
            0.024783754721283913
          ],
          "n": 8
        }
      }
    ],
    "track_justification": "dynamics is the built-in structural match for rollout failure/control states"
  }
}