- Reject malformed requests and sub-day cleanup windows
- Reset deduplication across media and subtitle-track changes
- Handle karaoke bursts ending with a long hold frame
- Drain the full write queue before scanning, not just one batch, so pending burst rows aren't missed
- Floor lookback-days before the positivity check so a sub-day value no longer collapses to a zero-day window
- Let the parsed cue list override the streaming dedup heuristic wherever it covers a line
- Reject combining --lifetime with --duplicate-lines and non-positive --lookback-days values
- Widen the burst frame bound to catch heavier typesetting while still sparing longer-spaced runs
- Keep the cleanup modal open until the user closes it so they can read the result before it unmounts
- Collapse animation-burst subtitle lines (karaoke OPs, animated signs) at ingest time using the same dedup rules the subtitle sidebar already applies, so repeated frames no longer flood "Top Repeated Words"
- Add retroactive cleanup for stats already affected: a "Duplicates" scanner/cleaner in the Vocabulary tab and `subminer stats cleanup --duplicate-lines` (`--dry-run`, `--lookback-days`) on the CLI
- Only subtitle lines and the vocabulary counts they feed are touched; watch time and lines-seen totals are left as recorded
sudacode
