Lab

Camcorder disc rescue

Done its job

Recovers home video from unfinalized DVD camcorder discs that the Mac sees as blank.

$ python3 scripts/map_disc.py --json-out disc.json

drive ready, media present

media DVD-R, 1x writable

mounted no (finalized: false)

track start blocks kind

1 0 16 blank

2 16 311872 video

3 311888 148224 video

4 460112 2097152 blank

2 video tracks, 0.94 GB to read

read-only: nothing is written to the disc

Step one: the mapper lists every track with its start block and length, and probes the first sector to tell video from blank. Mockup with sample data.

$ python3 scripts/extract_disc.py --map disc.json --out ~/Desktop/Trip

[1/2] track 2: 311872 sectors

########## 100% 2.6 MB/s

0 bad sectors [clean]

video 28,410 frames audio 947.9 s

recordings 6

[2/2] track 3: 148224 sectors

########## 100% 2.4 MB/s

32 bad sectors [tail only]

video 12,882 frames audio 429.8 s

recordings 3

damage_is_tail_only: true

report: Trip/extract_report.json

Step two: each track is read sector by sector, with bad sectors counted and checked for run-out at the end of a recording. Mockup with sample data.

$ python3 scripts/convert_clips.py --in ~/Desktop/Trip --crf 20

=== track_02: 720x480, 29.97 fps

clip_001.mp4 41 MB 02:31 frames ok

clip_002.mp4 96 MB 05:58 frames ok

clip_003.mp4 27 MB 01:44 frames ok

=== track_03: 720x480, 29.97 fps

clip_004.mp4 63 MB 03:52 frames ok

clip_005.mp4 38 MB 02:20 frames ok

verified 5 clips, 22:25 total

duration = frames / frame rate

Steps three and four: each recording becomes one MP4, and the verification line compares output frames with the frames read from the disc. Mockup with sample data.

How it works

Check the drive and the disc. drutil confirms the burner is attached and media is present, and a short wait avoids reading blank data from a disc that is still spinning up.
Map the disc. The mapper reports whether a filesystem exists, then lists every track with its start block, length and content type, probed from the first sector instead of guessed.
Extract the video objects. Each video track is read sector by sector into a folder I name first. Bad sectors are counted, and damage only at the end of a track is treated as harmless run-out.
Convert with ffmpeg. Each recording becomes one deinterlaced H.264 MP4, with an optional lossless MKV copy as an archive master.
Verify and report. Output frames are compared with the frames read from the disc, and the true duration comes from frames divided by frame rate.

Built with

PythonThree scripts: map the disc, extract the tracks, convert the clips.
ffmpegEncodes each recording to MP4 and optionally keeps a lossless MKV.
macOS drutilReports the drive, the media and the device node, read-only.
Claude Code skillPackages the workflow and its rules so an agent walks through it the same way each time.

Decisions

Never write to the disc. No burn, erase or finalize command is ever run. Closing the session would stop the camcorder from finalizing the disc properly later, and a read-only rip can always be retried.
Convert through raw streams. Camcorder files restart their timestamps at every recording, so a direct conversion either adds silence or drifts out of sync. Going through elementary streams and rebuilding the timing avoids both.
Trust frame counts, not file metadata. The duration a tool reports for a raw carve is wrong, so the report uses frames divided by frame rate and checks it against the audio length.