Stop silent data loss in voice capture pipeline

Empty transcripts and transcription failures previously
deleted the temp audio and returned without writing any
record to disk — violating parity-at-encode (raw content
is episodic context, not noise).

- Preserve audio in Journal/Media/YYYY-MM/ on all paths
  (success, empty, failure) instead of unlinking.
- Write a markdown entry to Journal/Captures/ on failure
  paths with status, audio_path, and error fields.
- Add status: saved to successful captures so frontmatter
  is uniform across success and failure.
- Fire SSE capture_saved events on all terminal paths,
  with status included.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

scripts/api.py

@@ -690,44 +690,92 @@ async def run_dreamer(request: Request, auth: str = Depends(require_auth)):
         return JSONResponse({"started": False, "error": str(e)})
 
 def transcribe_and_save(tmp_path, timestamp, nextcloud_url, nextcloud_user, nextcloud_password):
-    """Background task — transcribes audio and saves to Nextcloud after endpoint returns."""
+    """Background task — transcribes audio and saves to Nextcloud after endpoint returns.
+    Audio is preserved in Journal/Media/ on every terminal path; failed and empty-transcript
+    captures still produce a markdown record in Journal/Captures/ with a status field."""
     import requests as req_lib
     nc_auth = (nextcloud_user, nextcloud_password)
+    month_dir = timestamp[:7]
+    audio_ext = os.path.splitext(tmp_path)[1] or ".webm"
+    audio_filename = f"{timestamp}-voice{audio_ext}"
+    audio_relpath = f"Journal/Media/{month_dir}/{audio_filename}"
+
+    def archive_audio() -> bool:
+        try:
+            with open(tmp_path, "rb") as f:
+                audio_bytes = f.read()
+            media_parent = f"{nextcloud_url}/remote.php/dav/files/{nextcloud_user}/Journal/Media"
+            media_dir = f"{media_parent}/{month_dir}"
+            req_lib.request("MKCOL", media_parent, auth=nc_auth, timeout=10)
+            req_lib.request("MKCOL", media_dir, auth=nc_auth, timeout=10)
+            req_lib.put(f"{media_dir}/{audio_filename}", data=audio_bytes, auth=nc_auth, timeout=60)
+            return True
+        except Exception as e:
+            print(f"Audio archival failed for {timestamp}: {e}")
+            return False
+        finally:
+            if os.path.exists(tmp_path):
+                os.unlink(tmp_path)
+
+    def write_capture(filename: str, content_md: str, status: str):
+        captures_dir = f"{nextcloud_url}/remote.php/dav/files/{nextcloud_user}/Journal/Captures"
+        try:
+            req_lib.request("MKCOL", captures_dir, auth=nc_auth, timeout=10)
+            req_lib.put(f"{captures_dir}/{filename}", data=content_md.encode("utf-8"), auth=nc_auth, timeout=30)
+        except Exception as e:
+            print(f"Capture markdown write failed for {timestamp}: {e}")
+            return
+        try:
+            payload = {"type": "capture_saved", "filename": filename, "timestamp": timestamp, "status": status}
+            req_lib.post("http://localhost:8000/api/events/notify", json=payload, timeout=3)
+            req_lib.post("http://localhost:8000/api/captures/events/notify", json=payload, timeout=3)
+        except Exception:
+            pass
+
+    transcript = ""
+    transcribe_error = None
     try:
         segments, _ = whisper_model.transcribe(
             tmp_path, language="en", vad_filter=True, beam_size=1, initial_prompt=WHISPER_PROMPT
         )
         transcript = " ".join(s.text.strip() for s in segments).strip()
-        os.unlink(tmp_path)
-        if not transcript:
-            print(f"Async transcription empty for {timestamp} — nothing saved")
-            return
-        filename = f"{timestamp}-voice.md"
-        content_md = f"# Capture — {timestamp}\n\n**type:** voice\n**modality:** audio\n**status:** unprocessed\n\n---\n\n{transcript}\n"
-        captures_dir = f"{nextcloud_url}/remote.php/dav/files/{nextcloud_user}/Journal/Captures"
-        req_lib.request("MKCOL", captures_dir, auth=nc_auth, timeout=10)
-        url = f"{captures_dir}/{filename}"
-        req_lib.put(url, data=content_md.encode("utf-8"), auth=nc_auth, timeout=30)
-        print(f"Async transcription saved: {filename}")
-        # Notify SSE clients that transcription is complete
-        try:
-            import requests as _req
-            _req.post("http://localhost:8000/api/events/notify", json={
-                "type": "capture_saved",
-                "filename": filename,
-                "timestamp": timestamp,
-            }, timeout=3)
-            _req.post("http://localhost:8000/api/captures/events/notify", json={
-                "type": "capture_saved",
-                "filename": filename,
-                "timestamp": timestamp,
-            }, timeout=3)
-        except Exception:
-            pass
     except Exception as e:
-        if os.path.exists(tmp_path):
+        transcribe_error = str(e)
-            os.unlink(tmp_path)
+
-        print(f"Async transcription failed for {timestamp}: {e}")
+    audio_archived = archive_audio()
+    audio_line = f"**audio_path:** {audio_relpath}\n" if audio_archived else "**audio_archive_failed:** true\n"
+
+    if transcribe_error is not None:
+        filename = f"{timestamp}-voice-failed.md"
+        content_md = (
+            f"# Capture — {timestamp}\n\n"
+            f"**type:** voice\n**modality:** audio\n**status:** failed_transcription\n"
+            f"{audio_line}"
+            f"**error:** {transcribe_error}\n"
+        )
+        write_capture(filename, content_md, "failed_transcription")
+        print(f"Async transcription failed for {timestamp}: {transcribe_error}")
+        return
+
+    if not transcript:
+        filename = f"{timestamp}-voice-empty.md"
+        content_md = (
+            f"# Capture — {timestamp}\n\n"
+            f"**type:** voice\n**modality:** audio\n**status:** empty_transcript\n"
+            f"{audio_line}"
+        )
+        write_capture(filename, content_md, "empty_transcript")
+        print(f"Async transcription empty for {timestamp}: audio archived")
+        return
+
+    filename = f"{timestamp}-voice.md"
+    content_md = (
+        f"# Capture — {timestamp}\n\n"
+        f"**type:** voice\n**modality:** audio\n**status:** saved\n"
+        f"{audio_line}\n---\n\n{transcript}\n"
+    )
+    write_capture(filename, content_md, "saved")
+    print(f"Async transcription saved: {filename}")
 
 
 @app.post("/api/capture")
@@ -834,7 +882,7 @@ Keep the full description to 150-250 words. Do not speculate beyond what is visi
 
 **type:** {capture_type}
 **modality:** {modality}
-**status:** unprocessed
+**status:** saved
 **media:** {media_path}
 {f"**project:** {project}" if project else ""}