{
  "claim": "Predicted goal-distance preference can conflict with execution; shortlist ranking loses fidelity.",
  "populations": {
    "a": "One curated recorded LeWM pair, confirmation start 176",
    "bcd": "All 96 audit starts, 63 non-expert candidates per start"
  },
  "definitions": {
    "gap": "(d_success - d_failure)/(d_success + d_failure), minimum predicted RMS distance in each outcome group",
    "pair_error": "Mean across mixed-outcome starts of success/failure pair fraction with failure assigned strictly lower predicted cost; ties are not inversions",
    "spearman": "Predicted vs realized latent goal costs, not binary-success correlation",
    "b_summary": "All points, median and interquartile interval; no inferential CI",
    "c_counts": "LeWM 96/96/34 and TD-JEPA 96/95/29 eligible starts at K=63/16/4"
  },
  "margin_summary": [
    {
      "model": "LeWM",
      "starts": 96,
      "failed_preferred": 8,
      "median": -0.25085547911050055,
      "q25": -0.40510211683017644,
      "q75": -0.09497428552944047
    },
    {
      "model": "TD-JEPA",
      "starts": 96,
      "failed_preferred": 8,
      "median": -0.10664919736956605,
      "q25": -0.24454043753077043,
      "q75": -0.04535328037486606
    }
  ],
  "pair_error_summary": [
    {
      "model": "LeWM",
      "candidates": 63,
      "mixed_starts": 96,
      "mean_inversion_percent": 3.9540800490853076
    },
    {
      "model": "LeWM",
      "candidates": 16,
      "mixed_starts": 96,
      "mean_inversion_percent": 20.442315569268697
    },
    {
      "model": "LeWM",
      "candidates": 4,
      "mixed_starts": 34,
      "mean_inversion_percent": 49.019607843137244
    },
    {
      "model": "TD-JEPA",
      "candidates": 63,
      "mixed_starts": 96,
      "mean_inversion_percent": 3.8731391910488404
    },
    {
      "model": "TD-JEPA",
      "candidates": 16,
      "mixed_starts": 95,
      "mean_inversion_percent": 19.43618552500131
    },
    {
      "model": "TD-JEPA",
      "candidates": 4,
      "mixed_starts": 29,
      "mean_inversion_percent": 38.2183908045977
    }
  ],
  "marks": [
    {
      "model": "LeWM",
      "metric": "within_start_spearman_mean",
      "scope": "all_deployable",
      "value": 0.8952911529944317
    },
    {
      "model": "LeWM",
      "metric": "pooled_spearman",
      "scope": "all_deployable",
      "value": 0.9306370736300524
    },
    {
      "model": "LeWM",
      "metric": "within_start_spearman_mean",
      "scope": "top_16",
      "value": 0.6209847683206887
    },
    {
      "model": "LeWM",
      "metric": "pooled_spearman",
      "scope": "top_16",
      "value": 0.7740274669080177
    },
    {
      "model": "LeWM",
      "metric": "within_start_spearman_mean",
      "scope": "top_4",
      "value": 0.10833333333333332
    },
    {
      "model": "LeWM",
      "metric": "pooled_spearman",
      "scope": "top_4",
      "value": 0.5194680830885929
    },
    {
      "model": "TD-JEPA",
      "metric": "within_start_spearman_mean",
      "scope": "all_deployable",
      "value": 0.7982641658789958
    },
    {
      "model": "TD-JEPA",
      "metric": "pooled_spearman",
      "scope": "all_deployable",
      "value": 0.8498181061917158
    },
    {
      "model": "TD-JEPA",
      "metric": "within_start_spearman_mean",
      "scope": "top_16",
      "value": 0.560215315597642
    },
    {
      "model": "TD-JEPA",
      "metric": "pooled_spearman",
      "scope": "top_16",
      "value": 0.6825967452498853
    },
    {
      "model": "TD-JEPA",
      "metric": "within_start_spearman_mean",
      "scope": "top_4",
      "value": 0.12621070589600872
    },
    {
      "model": "TD-JEPA",
      "metric": "pooled_spearman",
      "scope": "top_4",
      "value": 0.35736371379965925
    }
  ]
}