fix: compare playlists, not durations, to detect a shifted enumeration #7

Merged

nugget merged 1 commit from fix/rip-disc-false-positive into main

2026-08-22 23:21:49 -05:00

claude commented

2026-08-21 15:42:40 -05:00

Collaborator

The duration check I added in #5 aborted a correct rip. On a season 7 disc:

### WRONG TITLE t3: expected 3521s, got 3523s
###   scan title 3 = 00800.mpls | rip title 3 = 00800.mpls
###   scan enumerated 14 titles, this rip enumerated 14
### ABORTING: every remaining index is suspect.

Same playlist, same enumeration — which is to say the right title. The scan
reports whole seconds while ffprobe measures the muxed stream, and a couple of
seconds of disagreement is routine. Having aborted, it skipped the disc's two
remaining titles, which had to be ripped again.

Duration was the wrong signal to lead with. The playlist each index resolves
to is the ground truth: a shifted enumeration means index N mapped to a
different playlist than the scan recorded for it. That is now the abort
condition; a duration disagreement is merely noted.

Verification

Both real cases, replayed from their actual logs:

Case	scan → rip	titles	verdict
S7D1 t3 — the correct rip it wrongly killed	`00800.mpls` → `00800.mpls`	14 vs 14	accept
S5D4 t3 — the genuine divergence from #5	`00800.mpls` → `00304.mpls`	29 vs 26	abort

So the check still catches the failure it was written for, and no longer fires
on measurement noise.

--tolerance now controls when a duration difference is reported rather than
when the run aborts, and defaults to 5s.

The duration check I added in #5 aborted a **correct** rip. On a season 7 disc: ``` ### WRONG TITLE t3: expected 3521s, got 3523s ### scan title 3 = 00800.mpls | rip title 3 = 00800.mpls ### scan enumerated 14 titles, this rip enumerated 14 ### ABORTING: every remaining index is suspect. ``` Same playlist, same enumeration — which is to say the right title. The scan reports whole seconds while `ffprobe` measures the muxed stream, and a couple of seconds of disagreement is routine. Having aborted, it skipped the disc's two remaining titles, which had to be ripped again. **Duration was the wrong signal to lead with.** The playlist each index resolves to is the ground truth: a shifted enumeration means index N mapped to a *different playlist* than the scan recorded for it. That is now the abort condition; a duration disagreement is merely noted. ### Verification Both real cases, replayed from their actual logs: | Case | scan → rip | titles | verdict | |---|---|---|---| | S7D1 t3 — the correct rip it wrongly killed | `00800.mpls` → `00800.mpls` | 14 vs 14 | **accept** | | S5D4 t3 — the genuine divergence from #5 | `00800.mpls` → `00304.mpls` | 29 vs 26 | **abort** | So the check still catches the failure it was written for, and no longer fires on measurement noise. `--tolerance` now controls when a duration difference is *reported* rather than when the run aborts, and defaults to 5s.

claude added 1 commit

2026-08-21 15:42:40 -05:00

fix: compare playlists, not durations, to detect a shifted enumeration bc3540bdd9

The duration check aborted a correct rip. On a season 7 disc it reported

    WRONG TITLE t3: expected 3521s, got 3523s
      scan title 3 = 00800.mpls | rip title 3 = 00800.mpls
      scan enumerated 14 titles, this rip enumerated 14

-- the same playlist under the same enumeration, which is to say the right
title. The scan reports whole seconds while ffprobe measures the muxed stream,
and roughly two seconds of disagreement is routine. Having aborted, it skipped
the disc's two remaining titles, which then had to be ripped again.

Duration was the wrong signal to lead with. The playlist each index resolves to
is the ground truth: a shifted enumeration means index N mapped to a DIFFERENT
playlist than the scan recorded. That is now the abort condition, and a
duration disagreement is merely noted.

Verified against both real cases: the season 7 rip is accepted (00800.mpls
both sides, 14 titles both sides), and the season 5 divergence still aborts
(00800.mpls vs 00304.mpls, 29 titles vs 26).