Fix: record_api_call once per query, not per key rotation

app.py

@@ -203,6 +203,9 @@ def call_gemini_api(prompt: str) -> str:
         logger.warning("RATE LIMIT EXCEEDED - returning fallback signal")
         return "RATE_LIMITED"
     
+    # Record ONE API call at the start (not per retry/rotation)
+    record_api_call()
+    
     if not GEMINI_API_KEYS:
         return "Please configure GEMINI_API_KEY for real responses."
     
@@ -215,9 +218,6 @@ def call_gemini_api(prompt: str) -> str:
         
         for attempt in range(max_retries_per_key):
             try:
-                # Record the call BEFORE making it (counts against limit)
-                record_api_call()
-                
                 response = requests.post(
                     url,
                     headers={"Content-Type": "application/json"},
@@ -238,7 +238,7 @@ def call_gemini_api(prompt: str) -> str:
                     return "No response generated."
                 
                 elif response.status_code in [400, 429, 403, 500, 502, 503]:
-                    # Bad request (invalid key), Rate limit, quota exceeded, or server error - try next key
+                    # Rate limit, quota exceeded, or server error - try next key
                     logger.warning(f"API key {current_key_index + 1} got {response.status_code}, rotating...")
                     get_next_api_key()
                     keys_tried += 1