Optimize: Lower Chunk Threshold to 0.90 to fix AI false negatives (relying on Heuristic Override for false positive protection)

app/core/detector.py

@@ -182,7 +182,10 @@ class DeepfakeDetector:
                     is_chunk_ai = False
                     if "fake" in lbl or "spoof" in lbl or "synthetic" in lbl:
                          # Strict threshold for indvidual chunks
-                         if scr >= 0.99:
+                         # REVISED (Phase 8): Lowered from 0.99 to 0.90.
+                         # 0.99 was too strict, causing false negatives for High-Quality AI (ElevenLabs).
+                         # We rely on Heuristic Override to protect noisy human files.
+                         if scr >= 0.90:
                              is_chunk_ai = True
                              chunk_score = scr
                          else: