{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "eta": 0.0
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "eta": 0.0
        },
        "mean": 0.13071522302925587
      },
      {
        "cfg": {
          "lr": 0.003,
          "eta": 0.0
        },
        "mean": 0.01457755290903151
      },
      {
        "cfg": {
          "lr": 0.006,
          "eta": 0.0
        },
        "mean": 0.006850881385616958
      }
    ],
    "full": {
      "mean": 0.006962434679735452,
      "std": 0.0022827002771013453,
      "per_seed": [
        0.007542507257312536,
        0.00671812891960144,
        0.006612597499042749,
        0.0065302918665111065,
        0.003842386417090893,
        0.005645386874675751,
        0.006438639014959335,
        0.012369539588689804
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.007029922096990049,
    "std": 0.002205360988149756,
    "per_seed": [
      0.007741106674075127,
      0.006892496719956398,
      0.006658521946519613,
      0.0070778545923531055,
      0.0037982785142958164,
      0.005655689164996147,
      0.0063418434001505375,
      0.012073585763573647
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 6.748741725459695e-05,
    "idea_wins": 3,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0001985994167625904,
      0.00017436780035495758,
      4.592444747686386e-05,
      0.000547562725841999,
      -4.410790279507637e-05,
      1.0302290320396423e-05,
      -9.679561480879784e-05,
      -0.00029595382511615753
    ],
    "p_value": 0.51155,
    "mde": 0.0002082852976156794,
    "mde_rel_pct": 2.991558372847722,
    "verdict": "no measurable effect",
    "system_worked": false
  },
  "mechanism_signature": {
    "math_check": {
      "gamma_G": 1.0,
      "L": 0.8,
      "threshold": 0.8,
      "rows": [
        {
          "eta": 0.5,
          "predicted_slope": 0.6000000000000001,
          "observed_slope": 0.5999100179959524,
          "relative_error": 0.00014997000674606592
        },
        {
          "eta": 1.0,
          "predicted_slope": -0.3999999999999999,
          "observed_slope": -0.40004000533411915,
          "relative_error": 0.00010001333529810477
        }
      ],
      "passed": true
    },
    "idea_grid": [
      {
        "cfg": {
          "lr": 0.006,
          "eta": 0.02
        },
        "result": {
          "mean": 0.007029922096990049,
          "std": 0.002205360988149756,
          "per_seed": [
            0.007741106674075127,
            0.006892496719956398,
            0.006658521946519613,
            0.0070778545923531055,
            0.0037982785142958164,
            0.005655689164996147,
            0.0063418434001505375,
            0.012073585763573647
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "eta": 0.08
        },
        "result": {
          "mean": 0.00731758784968406,
          "std": 0.002062769628747506,
          "per_seed": [
            0.00851514283567667,
            0.006746230181306601,
            0.0072161052376031876,
            0.009792596101760864,
            0.004097180441021919,
            0.005617870483547449,
            0.005928367376327515,
            0.010627210140228271
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.006,
          "eta": 0.2
        },
        "result": {
          "mean": 0.008025791961699724,
          "std": 0.0031914273028467188,
          "per_seed": [
            0.009828031063079834,
            0.006756746210157871,
            0.0079221585765481,
            0.015269015915691853,
            0.004710389766842127,
            0.005246929824352264,
            0.005789421033114195,
            0.00868364330381155
          ],
          "n": 8
        }
      }
    ],
    "trained_behavior": [
      {
        "seed": 0,
        "baseline_disagreement": 0.25682780146598816,
        "idea_disagreement": 0.24640725553035736
      },
      {
        "seed": 1,
        "baseline_disagreement": 0.23402994871139526,
        "idea_disagreement": 0.22542807459831238
      },
      {
        "seed": 2,
        "baseline_disagreement": 0.21436160802841187,
        "idea_disagreement": 0.2034529596567154
      },
      {
        "seed": 3,
        "baseline_disagreement": 0.24291306734085083,
        "idea_disagreement": 0.2349667102098465
      },
      {
        "seed": 4,
        "baseline_disagreement": 0.31340721249580383,
        "idea_disagreement": 0.3070146143436432
      },
      {
        "seed": 5,
        "baseline_disagreement": 0.24664676189422607,
        "idea_disagreement": 0.23655381798744202
      },
      {
        "seed": 6,
        "baseline_disagreement": 0.23078785836696625,
        "idea_disagreement": 0.21963292360305786
      },
      {
        "seed": 7,
        "baseline_disagreement": 0.24261848628520966,
        "idea_disagreement": 0.2282203882932663
      }
    ],
    "prediction": "negative disagreement trend when eta*gamma_G exceeds L",
    "predicted_threshold": 0.8,
    "observed_mean_baseline_disagreement": 0.2476990930736065,
    "observed_mean_idea_disagreement": 0.23770959302783012,
    "confirmed": true
  },
  "parameter_counts": {
    "baseline": 4049,
    "idea": 4049
  },
  "protocol_notes": "8 paired seeds; baseline sweep uses 4 seeds and full re-evaluation; n_train=n_test=400; same ParallelGRU base and optimizer."
}