Align, merge, and report overlapping DV tape captures — the same workflow as hdvmerge, but for DV instead of HDV. When a worn DV tape reads poorly you capture it in several passes — stop, rewind a little, retry whenever mosaic starts creeping in — ending up with several .dv files that overlap and each carry some errors. dvmerge folds them into one best-of file and tells you exactly which tape spots, if any, still need another pass.

It is a thin wrapper around dvrescue (MIPoPS), which already does the hard part for DV: it aligns every capture by the tape's absolute track number, picks each frame's cleanest copy across passes block by block, and writes a valid DV stream with its metadata intact. dvmerge drives one dvrescue merge and turns its per-frame CSV log into the re-capture list — the report you read each round.

A dirty tape doesn't fail all at once: it reads fine for a while, then mosaic builds up as debris loads onto the heads, and you can't watch it every second. So you re-capture blind — rewind a bit, go again — and hope the new pass caught what the last one smeared. dvmerge removes the guesswork: it compares every pass and prints the spots where no pass has a clean frame, each cued by tape timecode (to find it on the deck) and the camera's recording time (as a cross-check). You re-shoot those, drop the new file in, and the list shrinks.

DV capture over FireWire copies the tape bit-for-bit, and every DV frame carries its absolute track number on tape (abst) plus the camera's recording clock (rdt) and SMPTE timecode (tc). dvrescue uses abst as a frame-accurate, metadata-free coordinate to line all the captures up and pick the best copy of each frame. dvmerge reads the result:

   overlapping captures (.dv, read-only)
        │
      merge    one `dvrescue … -m merged.dv --merge-log log.csv --csv` run:
        │      aligns every capture by tape position, picks each frame's cleanest copy,
        │      writes the merged DV + a per-frame log. (Cached by input fingerprint.)
        ▼
      report   parse the log: every frame is clean / mosaic (present but still damaged) /
        │      missing (no capture has it). Coalesce the imperfect ones into re-capture spans,
        │      each tagged with tape TC, recording time, and how many captures cover it.
        └─►  with -o   keep merged.dv and write merged.dv.report.md beside it

The merge log is cached next to the captures (.dvmerge/), so re-reading the report or re-tuning --bridge is instant; adding or changing a capture re-runs the merge. Without -o, nothing large is kept — the merged DV goes to a temp file that is deleted once its log is parsed.

Python 3.9+ and the dvrescue CLI on your PATH. No third-party Python packages.

pip install -e .          # provides the `dvmerge` command

dvrescue is from MediaArea/MIPoPS (brew install dvrescue on macOS).

dvmerge CLIP-*.dv                  # analyse: merge, print the re-capture list
dvmerge CLIP-*.dv -o merged.dv     # same, then keep merged.dv and write merged.dv.report.md

The loop: run it to see what needs re-capturing → re-capture those tape spots (cue on the printed tape TC, leave ≥15 s of good footage on both sides, and overlap the previous good take so the passes join with no gap) → drop the new file in → run again until the list is only physically unreadable spots → add -o to keep the merged file.

• One best-of DV from many overlapping captures: dvrescue picks each tape frame's cleanest copy, block by block, across every pass — so a mosaic in one capture is repaired from another wherever a cleaner copy exists.
• A re-capture list: the exact tape spans where the merged frame is still imperfect, split into mosaic (present but damaged — you have dirty copies to improve on) and missing (no capture has it — content is absent until re-captured), each with tape TC, recording time, duration, severity, and which captures cover it.
• The merge, the merged DV, and its metadata are all dvrescue's — dvmerge never re-encodes.

• The re-capture report reflects what made it into the merged output. Frames missing from every capture have no row in dvrescue's log; dvmerge recovers them from jumps in the absolute track number abst (the physical tape position), so a camera stop/start or non-record-run timecode is not mistaken for missing tape — only a real gap in tape position counts. If a capture reports no abst at all, it falls back to the timecode delta there.
• --fps is a single value per run; mixing PAL and NTSC captures in one merge is not supported (nor would aligning them make sense).
• dvmerge delegates the merge entirely to dvrescue; its correctness and metadata handling are dvrescue's. dvmerge owns only discovery, caching, and the report.

python -m unittest discover -s tests        # deterministic; no dvrescue or sample captures needed

Tests build tiny synthetic merge logs in tests/test_plan.py.