SYNC_CHAPTERS fails with `ENOENT: no such file or directory, copyfile` pointing at a /tmp/storyteller-platform-align-<id> dir

I replaced the EPUB of a book that already had a readaloud and re-ran the alignment. Transcription finished, then the job died with ERROR: Encountered error while running task SYNC_CHAPTERS, ENOENT: no such file or directory, copyfile ..., referencing a /tmp/storyteller-platform-align-<id> directory that does not exist in the container (only unrelated /tmp/storyteller-align-silence-<id> dirs are there). Docker Desktop on Windows. Every retry failed the same way, and each attempt cost 10+ hours of transcription on my slow server.

Resolution: The underlying cause was the EPUB, not the container: the file was a single giant content document with no chapter splits, so chapter synchronization had nothing to map the audio onto and blew up in the copy step. Sourcing a properly split EPUB of the same book (147 content documents instead of 1) made alignment succeed against the exact same audio. To try different EPUBs without re-transcribing, reuse the transcripts: back up /data/assets/<Book>/transcriptions to the host; delete the readaloud from the web UI; remove the old .epub and _epub2.epub from the library folder and drop the new EPUB in; re-scan the library; start "Create readaloud"; then, while it is still pre-processing/transcoding the audio (a minute or two), copy the transcriptions folder back into /data/assets/<Book>/. The job sees the existing transcripts and skips transcription in seconds, going straight to alignment. Do not copy the folder back before starting the job -- the pipeline deletes transcoded audio/ and transcriptions/ at the start of the run, and pausing the run does not stop that. Also check the ratio of output size to source audio: an EPUB that only partially matches will produce a suspiciously small readaloud (e.g. 60 MB out of 780 MB of audio) because whole stretches of sentences get skipped.