fix: retry words request on 400/404 + add backend diagnostic logging

Frontend: retry /words POST once after 2s delay if it gets 400/404,
which happens when navigating via wizard after container restart
(session cache not yet warm).

Backend: log when session needs DB reload and when dewarped_bgr is missing.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

admin-lehrer/components/ocr-pipeline/StepWordRecognition.tsx

@@ -105,12 +105,24 @@ export function StepWordRecognition({ sessionId, onNext, goToStep }: StepWordRec
     setGridResult(null)
 
     try {
-      const res = await fetch(
+      // Retry once if initial request fails (e.g. after container restart,
-        `${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/words?stream=true&engine=${eng}&pronunciation=${pronunciation}`,
+      // session cache may not be warm yet when navigating via wizard)
-        { method: 'POST' },
+      let res: Response | null = null
-      )
+      for (let attempt = 0; attempt < 2; attempt++) {
-      if (!res.ok) {
+        res = await fetch(
-        const err = await res.json().catch(() => ({ detail: res.statusText }))
+          `${KLAUSUR_API}/api/v1/ocr-pipeline/sessions/${sessionId}/words?stream=true&engine=${eng}&pronunciation=${pronunciation}`,
+          { method: 'POST' },
+        )
+        if (res.ok) break
+        if (attempt === 0 && (res.status === 400 || res.status === 404)) {
+          // Wait briefly for cache to warm up, then retry
+          await new Promise(r => setTimeout(r, 2000))
+          continue
+        }
+        break
+      }
+      if (!res || !res.ok) {
+        const err = await res?.json().catch(() => ({ detail: res?.statusText })) || { detail: 'Worterkennung fehlgeschlagen' }
         throw new Error(err.detail || 'Worterkennung fehlgeschlagen')
       }
 

klausur-service/backend/ocr_pipeline_api.py

@@ -1204,11 +1204,14 @@ async def detect_words(
         stream: false (default) for JSON response, true for SSE streaming
     """
     if session_id not in _cache:
+        logger.info("detect_words: session %s not in cache, loading from DB", session_id)
         await _load_session_to_cache(session_id)
     cached = _get_cached(session_id)
 
     dewarped_bgr = cached.get("dewarped_bgr")
     if dewarped_bgr is None:
+        logger.warning("detect_words: dewarped_bgr is None for session %s (cache keys: %s)",
+                       session_id, [k for k in cached.keys() if k.endswith('_bgr')])
         raise HTTPException(status_code=400, detail="Dewarp must be completed before word detection")
 
     session = await get_session_db(session_id)