{
  "bench_version": 1,
  "track": "dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.006,
      "epochs": 15,
      "keep": 32,
      "method": "magnitude"
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "epochs": 15,
          "keep": 32,
          "method": "magnitude"
        },
        "mean": 0.0010415702417958528
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 15,
          "keep": 32,
          "method": "magnitude"
        },
        "mean": 0.0007543887477368116
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 15,
          "keep": 32,
          "method": "magnitude"
        },
        "mean": 0.0004434098445926793
      }
    ],
    "full": {
      "mean": 0.0004183898890914861,
      "std": 0.00018028736524803996,
      "per_seed": [
        0.0002669677196536213,
        0.00046067332732491195,
        0.0002578582789283246,
        0.0007881400524638593,
        0.0002825864066835493,
        0.0005642541800625622,
        0.0002470404433552176,
        0.00047959870425984263
      ],
      "n": 8
    }
  },
  "idea": {
    "best_cfg": {
      "lr": 0.006,
      "epochs": 15,
      "keep": 32,
      "method": "sensitivity"
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.0015,
          "epochs": 15,
          "keep": 32,
          "method": "sensitivity"
        },
        "mean": 0.0051941196143161505
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 15,
          "keep": 32,
          "method": "sensitivity"
        },
        "mean": 0.0029561036499217153
      },
      {
        "cfg": {
          "lr": 0.006,
          "epochs": 15,
          "keep": 32,
          "method": "sensitivity"
        },
        "mean": 0.0012721883977064863
      }
    ],
    "mean": 0.0012721883977064863,
    "std": 0.0006724726946615953,
    "per_seed": [
      0.0011970085324719548,
      0.0006662395317107439,
      0.0011209469521418214,
      0.0028435320127755404,
      0.001395156024955213,
      0.0005800570943392813,
      0.0015368196181952953,
      0.0008377474150620401
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.0008537985086150002,
    "idea_wins": 0,
    "n_pairs": 8,
    "per_seed_diffs": [
      0.0009300408128183335,
      0.00020556620438583195,
      0.0008630886732134968,
      0.002055391960311681,
      0.0011125696182716638,
      1.5802914276719093e-05,
      0.0012897791748400778,
      0.00035814871080219746
    ],
    "p_value": 0.0081,
    "mde": 0.000554180300336378,
    "mde_rel_pct": 132.45547150762997,
    "verdict": "idea worse (significant)",
    "system_worked": false
  },
  "mechanism_signature": {
    "prediction": "weighted sensitivity information and incremental residual rank identify useful hidden groups",
    "per_seed": [
      {
        "retained_units": 32,
        "params_total": 13313,
        "mean_information_retained": 9.77945613861084,
        "mean_residual_retained": 3.6348969729260716e-05,
        "importance_ablation_corr": 0.477768394227869,
        "predicted_information": 11.743715286254883,
        "observed_ablation": 0.000710174660306643
      },
      {
        "retained_units": 32,
        "params_total": 13313,
        "mean_information_retained": 9.776529312133789,
        "mean_residual_retained": 5.1056572374363896e-05,
        "importance_ablation_corr": 0.42190104414981555,
        "predicted_information": 11.318390846252441,
        "observed_ablation": 0.0006673403675832024
      },
      {
        "retained_units": 32,
        "params_total": 13313,
        "mean_information_retained": 9.697711944580078,
        "mean_residual_retained": 2.9330890470191662e-05,
        "importance_ablation_corr": 0.47564845830287067,
        "predicted_information": 10.515939712524414,
        "observed_ablation": 0.0006880304698516637
      },
      {
        "retained_units": 32,
        "params_total": 13313,
        "mean_information_retained": 8.47030258178711,
        "mean_residual_retained": 4.5788009657599105e-05,
        "importance_ablation_corr": 0.5436690158311026,
        "predicted_information": 10.03992748260498,
        "observed_ablation": 0.000801966964582998
      },
      {
        "retained_units": 32,
        "params_total": 13313,
        "mean_information_retained": 6.386395454406738,
        "mean_residual_retained": 1.4392690587783363e-05,
        "importance_ablation_corr": 0.18481644015160745,
        "predicted_information": 9.996113777160645,
        "observed_ablation": 0.0005640771121928623
      },
      {
        "retained_units": 32,
        "params_total": 13313,
        "mean_information_retained": 10.290732383728027,
        "mean_residual_retained": 3.206331243177374e-05,
        "importance_ablation_corr": 0.5280788239090402,
        "predicted_information": 11.632051467895508,
        "observed_ablation": 0.000697915483999556
      },
      {
        "retained_units": 32,
        "params_total": 13313,
        "mean_information_retained": 11.150772094726562,
        "mean_residual_retained": 5.351222108629372e-05,
        "importance_ablation_corr": 0.5366510030845261,
        "predicted_information": 12.404411315917969,
        "observed_ablation": 0.0007341111947982149
      },
      {
        "retained_units": 32,
        "params_total": 13313,
        "mean_information_retained": 9.854597091674805,
        "mean_residual_retained": 3.187904701462685e-05,
        "importance_ablation_corr": 0.38449025322316216,
        "predicted_information": 12.160406112670898,
        "observed_ablation": 0.000575491002093198
      }
    ],
    "confirmed": false
  },
  "protocol_notes": {
    "paired_seeds": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "n_train": 400,
    "n_test": 200,
    "retained_units": 32,
    "effective_unit_fraction": 0.5,
    "baseline_method": "outgoing-head magnitude selection",
    "idea_method": "trained-output sensitivity trace plus leave-one-group residual selection",
    "custom_track": null
  }
}