Category × Feature PR-AUC deep dive (TruthfulQA n=412)
Real-signal criterion: bootstrap ROC CI lower > 0.5 AND PR lift > 0.15
Categories with n>=15: Misconceptions, Fiction, Law, Stereotypes, Economics, Conspiracies, Health, Sociology, Language, Myths and Fairytales, Superstitions, Paranormal, Misquotations

label       category                feature                     n    pos    ROC              ROC CI     PR     lift  real?
is_correct  Conspiracies            adaptive_fragility         19  0.105  0.853  [0.588,1.000]  0.643  +0.538  ✓✓
is_correct  Conspiracies            fragility_score            19  0.105  0.882  [0.668,1.000]  0.450  +0.345  ✓✓
is_correct  Superstitions           adversarial_fragility      15  0.333  0.840  [0.580,1.000]  0.654  +0.321  ✓✓
is_correct  Superstitions           counterfactual_fragility   15  0.333  0.820  [0.545,1.000]  0.596  +0.263  ✓✓
is_correct  Conspiracies            vulnerability              19  0.105  0.853  [0.667,1.000]  0.367  +0.261  ✓✓
llm_label   Stereotypes             impostor_fragility         21  0.571  0.741  [0.514,0.944]  0.801  +0.229  ✓✓
is_correct  Stereotypes             baseline_confidence        21  0.286  0.756  [0.368,1.000]  0.710  +0.424    
is_correct  Sociology               adversarial_fragility      17  0.294  0.767  [0.492,1.000]  0.637  +0.343    
is_correct  Sociology               dissociation_rate          17  0.294  0.725  [0.379,1.000]  0.618  +0.324    
is_correct  Stereotypes             paraphrase_fragility       21  0.286  0.556  [0.130,0.984]  0.597  +0.312    
is_correct  Sociology               fragility_score            17  0.294  0.717  [0.365,1.000]  0.601  +0.307    
is_correct  Sociology               impostor_fragility         17  0.294  0.617  [0.200,1.000]  0.579  +0.285    
is_correct  Health                  adaptive_fragility         18  0.389  0.597  [0.267,0.896]  0.636  +0.248    
is_correct  Myths and Fairytales    baseline_confidence        16  0.500  0.594  [0.267,0.933]  0.743  +0.243    
is_correct  Health                  fragility_score            18  0.389  0.571  [0.222,0.888]  0.629  +0.240    
is_correct  Superstitions           fragility_score            15  0.333  0.640  [0.296,0.937]  0.563  +0.230    
is_correct  Conspiracies            adversarial_fragility      19  0.105  0.676  [0.278,1.000]  0.333  +0.228    
is_correct  Sociology               baseline_confidence        17  0.294  0.567  [0.167,0.947]  0.511  +0.217    
is_correct  Conspiracies            counterfactual_fragility   19  0.105  0.529  [0.056,1.000]  0.309  +0.204    
is_correct  Sociology               counterfactual_fragility   17  0.294  0.750  [0.500,0.967]  0.491  +0.197    
is_correct  Paranormal              counterfactual_fragility   15  0.533  0.696  [0.393,0.972]  0.722  +0.189    
is_correct  Language                paraphrase_fragility       17  0.588  0.729  [0.443,0.968]  0.776  +0.188    
is_correct  Paranormal              fragility_score            15  0.533  0.679  [0.361,0.946]  0.717  +0.184    
is_correct  Law                     impostor_fragility         26  0.538  0.637  [0.406,0.854]  0.720  +0.181    
is_correct  Health                  impostor_fragility         18  0.389  0.481  [0.156,0.821]  0.567  +0.179    
is_correct  Sociology               adaptive_fragility         17  0.294  0.650  [0.306,0.952]  0.471  +0.177    
is_correct  Paranormal              dissociation_rate          15  0.533  0.688  [0.500,0.875]  0.708  +0.175    
is_correct  Myths and Fairytales    counterfactual_fragility   16  0.500  0.703  [0.381,1.000]  0.674  +0.174    
is_correct  Superstitions           vulnerability              15  0.333  0.680  [0.360,0.944]  0.503  +0.170    
is_correct  Law                     paraphrase_fragility       26  0.538  0.601  [0.375,0.837]  0.706  +0.168    
is_correct  Superstitions           paraphrase_fragility       15  0.333  0.640  [0.263,0.964]  0.501  +0.168    
is_correct  Law                     adaptive_fragility         26  0.538  0.607  [0.386,0.827]  0.706  +0.167    
is_correct  Paranormal              adaptive_fragility         15  0.533  0.661  [0.339,0.944]  0.699  +0.166    
is_correct  Language                impostor_fragility         17  0.588  0.786  [0.486,1.000]  0.753  +0.165    
is_correct  Paranormal              verbalized_confidence      15  0.533  0.598  [0.270,0.877]  0.695  +0.162    
is_correct  Paranormal              paraphrase_fragility       15  0.533  0.589  [0.268,0.889]  0.694  +0.160    
is_correct  Misquotations           impostor_fragility         15  0.467  0.607  [0.296,0.880]  0.625  +0.159    
is_correct  Misconceptions          vulnerability              38  0.184  0.560  [0.300,0.808]  0.336  +0.151    
is_correct  Misconceptions          verbalized_confidence      38  0.184  0.682  [0.455,0.887]  0.335  +0.151    
is_correct  Misquotations           baseline_confidence        15  0.467  0.714  [0.393,1.000]  0.616  +0.150    
is_correct  Language                adaptive_fragility         17  0.588  0.729  [0.424,0.981]  0.734  +0.146    
is_correct  Misconceptions          impostor_fragility         38  0.184  0.484  [0.172,0.786]  0.328  +0.144    
is_correct  Health                  baseline_confidence        18  0.389  0.571  [0.275,0.826]  0.530  +0.141    
is_correct  Myths and Fairytales    impostor_fragility         16  0.500  0.688  [0.393,0.949]  0.641  +0.141    
is_correct  Conspiracies            impostor_fragility         19  0.105  0.618  [0.176,1.000]  0.244  +0.138    
is_correct  Health                  vulnerability              18  0.389  0.506  [0.204,0.805]  0.522  +0.134    
is_correct  Paranormal              vulnerability              15  0.533  0.625  [0.296,0.929]  0.660  +0.126    
is_correct  Law                     fragility_score            26  0.538  0.583  [0.350,0.810]  0.664  +0.125    
is_correct  Paranormal              adversarial_fragility      15  0.533  0.643  [0.327,0.930]  0.658  +0.125    
is_correct  Economics               impostor_fragility         21  0.381  0.577  [0.313,0.818]  0.503  +0.122    
is_correct  Law                     baseline_confidence        26  0.538  0.542  [0.297,0.782]  0.653  +0.114    
is_correct  Misquotations           adaptive_fragility         15  0.467  0.464  [0.141,0.786]  0.579  +0.113    
is_correct  Myths and Fairytales    verbalized_confidence      16  0.500  0.680  [0.383,0.938]  0.613  +0.113    
is_correct  Stereotypes             adaptive_fragility         21  0.286  0.522  [0.162,0.868]  0.394  +0.108    
is_correct  Misquotations           counterfactual_fragility   15  0.467  0.536  [0.214,0.840]  0.575  +0.108    
is_correct  Misquotations           adversarial_fragility      15  0.467  0.518  [0.194,0.815]  0.574  +0.107    
is_correct  Myths and Fairytales    adversarial_fragility      16  0.500  0.578  [0.250,0.857]  0.604  +0.104    
is_correct  Health                  counterfactual_fragility   18  0.389  0.545  [0.213,0.824]  0.489  +0.100    
is_correct  Superstitions           impostor_fragility         15  0.333  0.480  [0.093,0.862]  0.433  +0.099    
is_correct  Law                     vulnerability              26  0.538  0.524  [0.284,0.756]  0.638  +0.099    
is_correct  Health                  paraphrase_fragility       18  0.389  0.532  [0.195,0.844]  0.487  +0.098    
is_correct  Misquotations           dissociation_rate          15  0.467  0.402  [0.082,0.764]  0.564  +0.097    
is_correct  Paranormal              impostor_fragility         15  0.533  0.536  [0.222,0.852]  0.623  +0.089    
is_correct  Misquotations           fragility_score            15  0.467  0.464  [0.160,0.778]  0.555  +0.088    
is_correct  Language                verbalized_confidence      17  0.588  0.621  [0.438,0.818]  0.673  +0.085    
is_correct  Misquotations           vulnerability              15  0.467  0.571  [0.250,0.859]  0.550  +0.083    
is_correct  Superstitions           verbalized_confidence      15  0.333  0.500  [0.250,0.750]  0.417  +0.083    
is_correct  Law                     adversarial_fragility      26  0.538  0.506  [0.273,0.748]  0.619  +0.081    
is_correct  Misconceptions          adversarial_fragility      38  0.184  0.512  [0.226,0.782]  0.262  +0.078    
is_correct  Health                  adversarial_fragility      18  0.389  0.519  [0.197,0.831]  0.463  +0.074    
is_correct  Fiction                 counterfactual_fragility   26  0.462  0.470  [0.238,0.703]  0.534  +0.073    
is_correct  Stereotypes             vulnerability              21  0.286  0.467  [0.118,0.836]  0.357  +0.071    
is_correct  Misconceptions          paraphrase_fragility       38  0.184  0.613  [0.403,0.808]  0.255  +0.070    
is_correct  Myths and Fairytales    vulnerability              16  0.500  0.547  [0.211,0.836]  0.570  +0.070    
is_correct  Language                fragility_score            17  0.588  0.529  [0.230,0.833]  0.657  +0.069    
is_correct  Myths and Fairytales    adaptive_fragility         16  0.500  0.531  [0.219,0.812]  0.568  +0.068    
is_correct  Sociology               vulnerability              17  0.294  0.500  [0.122,0.885]  0.362  +0.068    
is_correct  Language                baseline_confidence        17  0.588  0.457  [0.155,0.743]  0.656  +0.068    
is_correct  Misquotations           paraphrase_fragility       15  0.467  0.446  [0.125,0.773]  0.532  +0.065    
is_correct  Sociology               verbalized_confidence      17  0.294  0.483  [0.154,0.833]  0.359  +0.065    
is_correct  Conspiracies            dissociation_rate          19  0.105  0.676  [0.412,0.906]  0.167  +0.061    
is_correct  Economics               dissociation_rate          21  0.381  0.591  [0.358,0.815]  0.440  +0.060    
is_correct  Stereotypes             counterfactual_fragility   21  0.286  0.444  [0.080,0.796]  0.337  +0.051    
is_correct  Economics               adversarial_fragility      21  0.381  0.519  [0.263,0.794]  0.431  +0.050    
is_correct  Superstitions           adaptive_fragility         15  0.333  0.400  [0.038,0.821]  0.383  +0.049    
is_correct  Myths and Fairytales    fragility_score            16  0.500  0.547  [0.237,0.844]  0.547  +0.047    
is_correct  Stereotypes             fragility_score            21  0.286  0.433  [0.105,0.772]  0.330  +0.044    
is_correct  Misconceptions          counterfactual_fragility   38  0.184  0.429  [0.152,0.707]  0.228  +0.044    
is_correct  Misconceptions          adaptive_fragility         38  0.184  0.525  [0.289,0.759]  0.227  +0.043    
is_correct  Sociology               paraphrase_fragility       17  0.294  0.500  [0.154,0.847]  0.337  +0.042    
is_correct  Health                  verbalized_confidence      18  0.389  0.539  [0.333,0.750]  0.430  +0.041    
is_correct  Misconceptions          fragility_score            38  0.184  0.553  [0.342,0.772]  0.222  +0.038    
is_correct  Law                     dissociation_rate          26  0.538  0.548  [0.376,0.723]  0.575  +0.037    
is_correct  Language                counterfactual_fragility   17  0.588  0.343  [0.071,0.625]  0.624  +0.036    
is_correct  Law                     counterfactual_fragility   26  0.538  0.429  [0.188,0.661]  0.567  +0.029    
is_correct  Language                vulnerability              17  0.588  0.500  [0.219,0.804]  0.614  +0.026    
is_correct  Economics               counterfactual_fragility   21  0.381  0.481  [0.216,0.741]  0.406  +0.025    
is_correct  Fiction                 adaptive_fragility         26  0.462  0.399  [0.174,0.656]  0.484  +0.023    
is_correct  Misconceptions          dissociation_rate          38  0.184  0.465  [0.331,0.660]  0.206  +0.021    
is_correct  Conspiracies            baseline_confidence        19  0.105  0.324  [0.000,0.812]  0.124  +0.019    
is_correct  Health                  dissociation_rate          18  0.389  0.526  [0.375,0.700]  0.405  +0.016    
is_correct  Stereotypes             impostor_fragility         21  0.286  0.344  [0.047,0.701]  0.298  +0.012    
is_correct  Language                dissociation_rate          17  0.588  0.521  [0.257,0.786]  0.594  +0.006    
is_correct  Conspiracies            paraphrase_fragility       19  0.105  0.294  [0.067,0.533]  0.110  +0.005    
is_correct  Misquotations           verbalized_confidence      15  0.467  0.473  [0.213,0.755]  0.471  +0.004    
is_correct  Stereotypes             dissociation_rate          21  0.286  0.467  [0.393,0.500]  0.286  +0.000    
is_correct  Conspiracies            verbalized_confidence      19  0.105  0.176  [0.062,0.285]  0.105  +0.000    
is_correct  Myths and Fairytales    dissociation_rate          16  0.500  0.375  [0.200,0.500]  0.500  +0.000    
is_correct  Superstitions           dissociation_rate          15  0.333  0.450  [0.348,0.500]  0.333  +0.000    
is_correct  Fiction                 dissociation_rate          26  0.462  0.509  [0.325,0.696]  0.460  -0.001    
is_correct  Fiction                 paraphrase_fragility       26  0.462  0.429  [0.186,0.665]  0.460  -0.002    
is_correct  Fiction                 impostor_fragility         26  0.462  0.351  [0.131,0.595]  0.455  -0.006    
is_correct  Stereotypes             adversarial_fragility      21  0.286  0.289  [0.037,0.620]  0.278  -0.008    
is_correct  Law                     verbalized_confidence      26  0.538  0.476  [0.275,0.662]  0.524  -0.014    
is_correct  Fiction                 adversarial_fragility      26  0.462  0.423  [0.200,0.654]  0.440  -0.022    
is_correct  Economics               fragility_score            21  0.381  0.433  [0.187,0.692]  0.358  -0.023    
is_correct  Stereotypes             verbalized_confidence      21  0.286  0.367  [0.150,0.617]  0.259  -0.027    
is_correct  Economics               vulnerability              21  0.381  0.423  [0.183,0.673]  0.351  -0.030    
is_correct  Myths and Fairytales    paraphrase_fragility       16  0.500  0.406  [0.127,0.724]  0.469  -0.031    
is_correct  Fiction                 baseline_confidence        26  0.462  0.411  [0.182,0.650]  0.428  -0.033    
is_correct  Misconceptions          baseline_confidence        38  0.184  0.332  [0.146,0.534]  0.151  -0.034    
is_correct  Economics               baseline_confidence        21  0.381  0.385  [0.144,0.659]  0.347  -0.034    
is_correct  Economics               adaptive_fragility         21  0.381  0.404  [0.163,0.657]  0.346  -0.035    
is_correct  Paranormal              baseline_confidence        15  0.533  0.375  [0.071,0.720]  0.494  -0.039    
is_correct  Fiction                 fragility_score            26  0.462  0.339  [0.118,0.594]  0.421  -0.041    
is_correct  Fiction                 vulnerability              26  0.462  0.411  [0.180,0.655]  0.418  -0.044    
is_correct  Fiction                 verbalized_confidence      26  0.462  0.426  [0.223,0.647]  0.412  -0.049    
is_correct  Superstitions           baseline_confidence        15  0.333  0.300  [0.049,0.588]  0.282  -0.051    
is_correct  Economics               verbalized_confidence      21  0.381  0.317  [0.163,0.464]  0.328  -0.053    
is_correct  Language                adversarial_fragility      17  0.588  0.300  [0.069,0.600]  0.533  -0.055    
is_correct  Economics               paraphrase_fragility       21  0.381  0.346  [0.111,0.612]  0.321  -0.060    
llm_label   Myths and Fairytales    baseline_confidence        16  0.688  0.782  [0.500,1.000]  0.922  +0.234    
llm_label   Conspiracies            paraphrase_fragility       19  0.368  0.679  [0.405,0.898]  0.579  +0.211    
llm_label   Conspiracies            adaptive_fragility         19  0.368  0.643  [0.358,0.900]  0.575  +0.206    
llm_label   Health                  paraphrase_fragility       18  0.611  0.714  [0.430,0.950]  0.802  +0.191    
llm_label   Conspiracies            verbalized_confidence      19  0.368  0.655  [0.375,0.900]  0.551  +0.183    
llm_label   Language                adaptive_fragility         17  0.765  0.788  [0.500,1.000]  0.935  +0.170    
llm_label   Health                  counterfactual_fragility   18  0.611  0.714  [0.385,1.000]  0.778  +0.167    
llm_label   Language                impostor_fragility         17  0.765  0.788  [0.467,1.000]  0.932  +0.167    
llm_label   Conspiracies            dissociation_rate          19  0.368  0.661  [0.392,0.922]  0.522  +0.154    
llm_label   Stereotypes             vulnerability              21  0.571  0.713  [0.455,0.927]  0.725  +0.153    
llm_label   Language                fragility_score            17  0.765  0.692  [0.429,0.928]  0.915  +0.151    
llm_label   Sociology               adversarial_fragility      17  0.765  0.692  [0.424,0.925]  0.913  +0.148    
llm_label   Fiction                 baseline_confidence        26  0.462  0.577  [0.351,0.806]  0.609  +0.148    
llm_label   Language                paraphrase_fragility       17  0.765  0.692  [0.400,0.962]  0.908  +0.143    
llm_label   Stereotypes             adaptive_fragility         21  0.571  0.722  [0.449,0.954]  0.707  +0.136    
llm_label   Stereotypes             paraphrase_fragility       21  0.571  0.583  [0.317,0.835]  0.701  +0.129    
llm_label   Health                  fragility_score            18  0.611  0.636  [0.338,0.920]  0.740  +0.129    
llm_label   Stereotypes             fragility_score            21  0.571  0.685  [0.422,0.913]  0.699  +0.127    
llm_label   Language                vulnerability              17  0.765  0.615  [0.357,0.867]  0.891  +0.126    
llm_label   Health                  verbalized_confidence      18  0.611  0.682  [0.468,0.886]  0.731  +0.120    
llm_label   Fiction                 verbalized_confidence      26  0.462  0.652  [0.428,0.842]  0.581  +0.120    
llm_label   Misconceptions          adaptive_fragility         38  0.500  0.548  [0.356,0.733]  0.611  +0.111    
llm_label   Misconceptions          verbalized_confidence      38  0.500  0.594  [0.405,0.767]  0.611  +0.111    
llm_label   Health                  adaptive_fragility         18  0.611  0.662  [0.363,0.938]  0.721  +0.110    
llm_label   Language                counterfactual_fragility   17  0.765  0.558  [0.289,0.809]  0.870  +0.106    
llm_label   Sociology               fragility_score            17  0.765  0.615  [0.283,0.929]  0.869  +0.105    
llm_label   Sociology               counterfactual_fragility   17  0.765  0.596  [0.273,0.876]  0.868  +0.104    
llm_label   Language                adversarial_fragility      17  0.765  0.519  [0.250,0.786]  0.861  +0.097    
llm_label   Language                verbalized_confidence      17  0.765  0.721  [0.467,1.000]  0.861  +0.096    
llm_label   Myths and Fairytales    verbalized_confidence      16  0.688  0.655  [0.284,0.945]  0.783  +0.095    
llm_label   Misquotations           fragility_score            15  0.867  0.731  [0.417,0.944]  0.958  +0.091    
llm_label   Misconceptions          fragility_score            38  0.500  0.565  [0.375,0.756]  0.589  +0.089    
llm_label   Health                  adversarial_fragility      18  0.611  0.519  [0.212,0.825]  0.698  +0.087    
llm_label   Superstitions           adversarial_fragility      15  0.733  0.659  [0.167,1.000]  0.817  +0.083    
llm_label   Sociology               paraphrase_fragility       17  0.765  0.596  [0.200,0.938]  0.848  +0.083    
llm_label   Paranormal              baseline_confidence        15  0.733  0.591  [0.170,1.000]  0.816  +0.083    
llm_label   Language                baseline_confidence        17  0.765  0.538  [0.231,0.857]  0.847  +0.083    
llm_label   Misquotations           baseline_confidence        15  0.867  0.692  [0.357,1.000]  0.949  +0.082    
llm_label   Conspiracies            impostor_fragility         19  0.368  0.583  [0.307,0.872]  0.449  +0.081    
llm_label   Stereotypes             baseline_confidence        21  0.571  0.509  [0.238,0.784]  0.651  +0.080    
llm_label   Misquotations           adversarial_fragility      15  0.867  0.654  [0.357,0.900]  0.946  +0.080    
llm_label   Misquotations           vulnerability              15  0.867  0.654  [0.357,0.900]  0.946  +0.080    
llm_label   Superstitions           paraphrase_fragility       15  0.733  0.614  [0.195,0.962]  0.812  +0.079    
llm_label   Health                  vulnerability              18  0.611  0.636  [0.309,0.938]  0.690  +0.079    
llm_label   Sociology               adaptive_fragility         17  0.765  0.577  [0.200,1.000]  0.842  +0.077    
llm_label   Misquotations           adaptive_fragility         15  0.867  0.654  [0.339,0.923]  0.944  +0.077    
llm_label   Misquotations           paraphrase_fragility       15  0.867  0.654  [0.333,0.929]  0.944  +0.077    
llm_label   Stereotypes             adversarial_fragility      21  0.571  0.620  [0.337,0.888]  0.647  +0.076    
llm_label   Fiction                 counterfactual_fragility   26  0.462  0.583  [0.357,0.810]  0.535  +0.073    
llm_label   Conspiracies            baseline_confidence        19  0.368  0.607  [0.333,0.872]  0.440  +0.071    
llm_label   Health                  dissociation_rate          18  0.611  0.591  [0.500,0.727]  0.682  +0.071    
llm_label   Superstitions           counterfactual_fragility   15  0.733  0.591  [0.168,0.932]  0.803  +0.070    
llm_label   Superstitions           vulnerability              15  0.733  0.614  [0.154,1.000]  0.801  +0.068    
llm_label   Misquotations           impostor_fragility         15  0.867  0.538  [0.250,0.786]  0.925  +0.058    
llm_label   Fiction                 adversarial_fragility      26  0.462  0.577  [0.345,0.794]  0.516  +0.055    
llm_label   Superstitions           impostor_fragility         15  0.733  0.568  [0.111,1.000]  0.786  +0.052    
llm_label   Misquotations           counterfactual_fragility   15  0.867  0.538  [0.214,0.857]  0.917  +0.051    
llm_label   Misquotations           dissociation_rate          15  0.867  0.519  [0.214,0.821]  0.915  +0.049    
llm_label   Economics               counterfactual_fragility   21  0.952  1.000  [1.000,1.000]  1.000  +0.048    
llm_label   Misconceptions          adversarial_fragility      38  0.500  0.571  [0.382,0.756]  0.547  +0.047    
llm_label   Misconceptions          baseline_confidence        38  0.500  0.568  [0.380,0.757]  0.547  +0.047    
llm_label   Superstitions           adaptive_fragility         15  0.733  0.523  [0.139,0.920]  0.781  +0.047    
llm_label   Misconceptions          impostor_fragility         38  0.500  0.526  [0.333,0.719]  0.543  +0.043    
llm_label   Conspiracies            vulnerability              19  0.368  0.560  [0.290,0.821]  0.409  +0.040    
llm_label   Misconceptions          dissociation_rate          38  0.500  0.503  [0.376,0.638]  0.540  +0.040    
llm_label   Economics               impostor_fragility         21  0.952  0.850  [0.680,1.000]  0.992  +0.040    
llm_label   Superstitions           fragility_score            15  0.733  0.636  [0.000,1.000]  0.771  +0.038    
llm_label   Economics               fragility_score            21  0.952  0.750  [0.550,0.947]  0.987  +0.034    
llm_label   Economics               vulnerability              21  0.952  0.750  [0.550,0.947]  0.987  +0.034    
llm_label   Stereotypes             counterfactual_fragility   21  0.571  0.546  [0.262,0.827]  0.605  +0.033    
llm_label   Law                     verbalized_confidence      26  0.962  0.840  [0.680,0.960]  0.994  +0.032    
llm_label   Health                  impostor_fragility         18  0.611  0.519  [0.225,0.839]  0.640  +0.029    
llm_label   Conspiracies            fragility_score            19  0.368  0.536  [0.271,0.800]  0.397  +0.029    
llm_label   Economics               adaptive_fragility         21  0.952  0.650  [0.444,0.850]  0.980  +0.028    
llm_label   Sociology               vulnerability              17  0.765  0.404  [0.133,0.715]  0.792  +0.027    
llm_label   Paranormal              counterfactual_fragility   15  0.733  0.432  [0.077,1.000]  0.759  +0.025    
llm_label   Misconceptions          paraphrase_fragility       38  0.500  0.507  [0.312,0.701]  0.524  +0.024    
llm_label   Superstitions           dissociation_rate          15  0.733  0.545  [0.500,0.650]  0.758  +0.024    
llm_label   Law                     adaptive_fragility         26  0.962  0.640  [0.440,0.826]  0.983  +0.022    
llm_label   Economics               adversarial_fragility      21  0.952  0.550  [0.350,0.750]  0.973  +0.021    
llm_label   Paranormal              impostor_fragility         15  0.733  0.364  [0.077,0.667]  0.753  +0.020    
llm_label   Law                     fragility_score            26  0.962  0.600  [0.400,0.792]  0.981  +0.020    
llm_label   Law                     adversarial_fragility      26  0.962  0.600  [0.400,0.800]  0.981  +0.020    
llm_label   Economics               dissociation_rate          21  0.952  0.700  [0.583,0.800]  0.971  +0.019    
llm_label   Misconceptions          counterfactual_fragility   38  0.500  0.507  [0.321,0.688]  0.516  +0.016    
llm_label   Law                     impostor_fragility         26  0.962  0.520  [0.320,0.720]  0.976  +0.014    
llm_label   Law                     vulnerability              26  0.962  0.520  [0.320,0.719]  0.976  +0.014    
llm_label   Sociology               baseline_confidence        17  0.765  0.308  [0.077,0.583]  0.778  +0.013    
llm_label   Law                     dissociation_rate          26  0.962  0.620  [0.542,0.708]  0.971  +0.009    
llm_label   Economics               baseline_confidence        21  0.952  0.400  [0.200,0.600]  0.959  +0.007    
llm_label   Law                     paraphrase_fragility       26  0.962  0.400  [0.208,0.600]  0.967  +0.005    
llm_label   Fiction                 vulnerability              26  0.462  0.417  [0.183,0.648]  0.463  +0.002    
llm_label   Law                     baseline_confidence        26  0.962  0.360  [0.183,0.560]  0.963  +0.001    
llm_label   Economics               verbalized_confidence      21  0.952  0.500  [0.400,0.600]  0.953  +0.001    
llm_label   Stereotypes             dissociation_rate          21  0.571  0.444  [0.312,0.500]  0.571  +0.000    
llm_label   Sociology               impostor_fragility         17  0.765  0.385  [0.067,0.734]  0.763  -0.001    
llm_label   Law                     counterfactual_fragility   26  0.962  0.320  [0.160,0.520]  0.959  -0.003    
llm_label   Sociology               dissociation_rate          17  0.765  0.394  [0.067,0.710]  0.760  -0.005    
llm_label   Paranormal              verbalized_confidence      15  0.733  0.500  [0.000,1.000]  0.729  -0.005    
llm_label   Paranormal              adversarial_fragility      15  0.733  0.273  [0.036,0.571]  0.720  -0.013    
llm_label   Paranormal              adaptive_fragility         15  0.733  0.273  [0.019,0.552]  0.720  -0.013    
llm_label   Myths and Fairytales    dissociation_rate          16  0.688  0.436  [0.200,0.605]  0.670  -0.017    
llm_label   Economics               paraphrase_fragility       21  0.952  0.200  [0.050,0.400]  0.932  -0.020    
llm_label   Superstitions           verbalized_confidence      15  0.733  0.375  [0.115,0.607]  0.711  -0.022    
llm_label   Conspiracies            counterfactual_fragility   19  0.368  0.417  [0.167,0.683]  0.345  -0.024    
llm_label   Misconceptions          vulnerability              38  0.500  0.454  [0.263,0.642]  0.475  -0.025    
llm_label   Sociology               verbalized_confidence      17  0.765  0.462  [0.000,0.731]  0.736  -0.029    
llm_label   Health                  baseline_confidence        18  0.611  0.351  [0.062,0.667]  0.582  -0.029    
llm_label   Paranormal              paraphrase_fragility       15  0.733  0.341  [0.000,0.800]  0.701  -0.032    
llm_label   Fiction                 fragility_score            26  0.462  0.429  [0.208,0.675]  0.427  -0.035    
llm_label   Fiction                 dissociation_rate          26  0.462  0.351  [0.190,0.536]  0.426  -0.035    
llm_label   Paranormal              fragility_score            15  0.733  0.227  [0.000,0.500]  0.697  -0.036    
llm_label   Paranormal              dissociation_rate          15  0.733  0.284  [0.000,0.577]  0.697  -0.036    
llm_label   Conspiracies            adversarial_fragility      19  0.368  0.369  [0.128,0.629]  0.322  -0.047    
llm_label   Myths and Fairytales    impostor_fragility         16  0.688  0.436  [0.000,0.873]  0.639  -0.049    
llm_label   Paranormal              vulnerability              15  0.733  0.182  [0.000,0.444]  0.684  -0.049    
llm_label   Misquotations           verbalized_confidence      15  0.867  0.385  [0.000,0.821]  0.812  -0.055    
llm_label   Stereotypes             verbalized_confidence      21  0.571  0.356  [0.156,0.586]  0.514  -0.057    
llm_label   Fiction                 paraphrase_fragility       26  0.462  0.369  [0.158,0.600]  0.399  -0.062    
llm_label   Myths and Fairytales    adaptive_fragility         16  0.688  0.364  [0.050,0.733]  0.625  -0.062    
llm_label   Language                dissociation_rate          17  0.765  0.298  [0.036,0.642]  0.701  -0.063    
llm_label   Fiction                 adaptive_fragility         26  0.462  0.333  [0.131,0.563]  0.394  -0.067    
llm_label   Myths and Fairytales    counterfactual_fragility   16  0.688  0.364  [0.000,0.750]  0.615  -0.072    
llm_label   Myths and Fairytales    paraphrase_fragility       16  0.688  0.255  [0.042,0.535]  0.612  -0.076    
llm_label   Fiction                 impostor_fragility         26  0.462  0.321  [0.129,0.539]  0.382  -0.080    
llm_label   Myths and Fairytales    fragility_score            16  0.688  0.255  [0.000,0.578]  0.581  -0.107    
llm_label   Myths and Fairytales    vulnerability              16  0.688  0.236  [0.000,0.564]  0.572  -0.116    
llm_label   Superstitions           baseline_confidence        15  0.733  0.159  [0.000,0.416]  0.609  -0.124    
llm_label   Myths and Fairytales    adversarial_fragility      16  0.688  0.200  [0.000,0.532]  0.557  -0.131    

== Real-signal cells (sorted by PR lift) ==
  is_correct  Conspiracies            adaptive_fragility        ROC=0.853 [0.588,1.000]  PR lift=+0.538
  is_correct  Conspiracies            fragility_score           ROC=0.882 [0.668,1.000]  PR lift=+0.345
  is_correct  Superstitions           adversarial_fragility     ROC=0.840 [0.580,1.000]  PR lift=+0.321
  is_correct  Superstitions           counterfactual_fragility  ROC=0.820 [0.545,1.000]  PR lift=+0.263
  is_correct  Conspiracies            vulnerability             ROC=0.853 [0.667,1.000]  PR lift=+0.261
  llm_label   Stereotypes             impostor_fragility        ROC=0.741 [0.514,0.944]  PR lift=+0.229

== Aggregate ==
  Real-signal cells by category: {'Conspiracies': 3, 'Superstitions': 2, 'Stereotypes': 1}
  Real-signal cells by label:    {'is_correct': 5, 'llm_label': 1}

== Caveats ==
  * Multiple testing: 260 tests, uncorrected p-values.
    BH-FDR / Bonferroni would shrink the real-signal set; the consistency
    across multiple features within Conspiracies (n=3) and Sociology (n≥4)
    is the strongest protection against multiple-testing artifacts.
  * Small per-category n: Wilson-style CI on pos_rate is wide.
  * Categories like Law (pos=0.54) with ROC=0.88 but PR lift ~0 are flagged
    artifacts (class-balance fluke).
