Update app.py

app.py

@@ -49,6 +49,9 @@ ESCALATION_QUESTIONS = [
     ("Violence has increased in frequency or severity", 3),
     ("Partner monitors your calls/GPS/social media", 2)
 ]
+DARVO_PATTERNS = {
+    "blame shifting", "projection", "dismissiveness", "guilt tripping", "contradictory statements"
+}
 DARVO_MOTIFS = [
     "I never said that.", "You’re imagining things.", "That never happened.",
     "You’re making a big deal out of nothing.", "It was just a joke.", "You’re too sensitive.",
@@ -180,13 +183,7 @@ def analyze_single_message(text, thresholds):
     }
 
     contradiction_flag = detect_contradiction(text)
-    motifs = [phrase for _, phrase in matched_phrases]
-    darvo_score = calculate_darvo_score(
-        pattern_labels,
-        sentiment_before=0.0,
-        sentiment_after=sentiment_score,
-        motifs_found=motifs,
-        contradiction_flag=contradiction_flag
+   
 )
     inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
     with torch.no_grad():
@@ -197,6 +194,13 @@ def analyze_single_message(text, thresholds):
         label for label, score in zip(LABELS, scores)
         if score > adjusted_thresholds[label]
     ]
+    motifs = [phrase for _, phrase in matched_phrases]
+    darvo_score = calculate_darvo_score(
+        pattern_labels,
+        sentiment_before=0.0,
+        sentiment_after=sentiment_score,
+        motifs_found=motifs,
+        contradiction_flag=contradiction_flag
 
     top_patterns = sorted(
         [(label, score) for label, score in zip(LABELS, scores)],
@@ -254,9 +258,6 @@ def analyze_composite(msg1, msg2, msg3, *answers_and_none):
     most_common_stage = max(set(stages), key=stages.count)
     stage_text = RISK_STAGE_LABELS[most_common_stage]
     avg_darvo = round(sum(darvo_scores) / len(darvo_scores), 3)
-    if avg_darvo > 0.25:
-        level = "moderate" if avg_darvo < 0.65 else "high"
-        out += f"\n\n🎭 **DARVO Score: {avg_darvo}** → This indicates a **{level} likelihood** of narrative reversal (DARVO), where the speaker may be denying, attacking, or reversing blame."
     top_label = f"{top_labels[0]} – {int(round(top_scores[0] * 100))}%"
     composite_abuse = int(round(sum(abuse_scores) / len(abuse_scores)))
 
@@ -268,7 +269,7 @@ def analyze_composite(msg1, msg2, msg3, *answers_and_none):
     out += f"\n\n{stage_text}"
 
     avg_darvo = round(sum(darvo_scores) / len(darvo_scores), 3)
-        if avg_darvo > 0.25:
+    if avg_darvo > 0.25:
         level = "moderate" if avg_darvo < 0.65 else "high"
         out += f"\n\n🎭 **DARVO Score: {avg_darvo}** → This indicates a **{level} likelihood** of narrative reversal (DARVO), where the speaker may be denying, attacking, or reversing blame."