Upload app.py

app.py

@@ -300,11 +300,14 @@ def predict_url(payload: PredictUrlPayload):
             "phish_is_positive_env": URL_POSITIVE_CLASS_ENV if URL_POSITIVE_CLASS_ENV else None,
         }
 
+        raw_p_class1_debug: Optional[float] = None
+
         if isinstance(model_type, str) and model_type == "xgboost_bst":
             if xgb is None:
                 raise RuntimeError("xgboost is not installed but required for this model bundle.")
             dmat = xgb.DMatrix(feats)
             raw_p_class1 = float(model.predict(dmat)[0])  # P(class == 1)
+            raw_p_class1_debug = raw_p_class1
             phish_proba = raw_p_class1 if phish_is_positive else (1.0 - raw_p_class1)
 
         elif hasattr(model, "predict_proba"):
@@ -348,9 +351,13 @@ def predict_url(payload: PredictUrlPayload):
         phish_proba = float(phish_proba)
         label = "PHISH" if phish_proba >= 0.5 else "LEGIT"
         score = phish_proba if label == "PHISH" else (1.0 - phish_proba)
+        # Map to numeric dataset-style label using resolved polarity
+        # If PHISH is the positive (class 1), PHISH -> 1 else 0; if not, invert
+        predicted_label_numeric = 1 if ((label == "PHISH") == bool(phish_is_positive)) else 0
 
         return {
             "label": label,
+            "predicted_label": int(predicted_label_numeric),
             "score": float(score),
             "phishing_probability": float(phish_proba),
             "backend": str(model_type),
@@ -359,6 +366,7 @@ def predict_url(payload: PredictUrlPayload):
             "phish_is_positive": bool(phish_is_positive),
             "phish_is_positive_bundle": meta_phish_is_positive,
             "phish_is_positive_env": URL_POSITIVE_CLASS_ENV if URL_POSITIVE_CLASS_ENV else None,
+            "raw_proba_class1": float(raw_p_class1_debug) if raw_p_class1_debug is not None else None,
             "feature_cols": feature_cols,
             "url_col": url_col,
         }