{
  "n_test": 96,
  "split": "campaign-held-out",
  "never_outputs_safe": true,
  "evidence_link_rate": 1.0,
  "hard_negative_suspicious_rate": 0.0,
  "decision_counts": {
    "suspicious": 48,
    "needs_review": 16,
    "insufficient_evidence": 32
  },
  "methods": {
    "rule-structural": {
      "pr_auc": 1.0,
      "brier": 0.10375000000000001,
      "recall_at_fpr_0_05": 1.0,
      "recall_at_fpr_0_10": 1.0,
      "mean_score": 0.34166666666666673,
      "pr_curve_points": 6
    },
    "tfidf-lr": {
      "pr_auc": 1.0,
      "brier": 0.14890835989583331,
      "recall_at_fpr_0_05": 1.0,
      "recall_at_fpr_0_10": 1.0,
      "mean_score": 0.629509375,
      "pr_curve_points": 43
    },
    "lightgbm-structural": {
      "pr_auc": 1.0,
      "brier": 0.16022288833333334,
      "recall_at_fpr_0_05": 1.0,
      "recall_at_fpr_0_10": 1.0,
      "mean_score": 0.7102166666666667,
      "pr_curve_points": 6
    }
  },
  "by_language": {
    "en": {
      "n": 64,
      "pr_auc_lightgbm": 1.0,
      "recall_at_fpr_0_10": 1.0
    },
    "fa": {
      "n": 32,
      "pr_auc_lightgbm": 1.0,
      "recall_at_fpr_0_10": 1.0
    }
  },
  "limitations": [
    "Synthetic fixtures only. Lab PR-AUC is not operational phishing accuracy.",
    "Copied Authentication-Results headers are not treated as proof.",
    "URLs are parsed and defanged; they are never fetched.",
    "A similar logo or a new domain is not enough for a malice verdict."
  ]
}