Introducing Version Control — QC Across Revisions
Post-production rarely stops at one pass. A colour grade changes, a subtitle gets fixed, the first two seconds get cut. Each revision needs a clean QC run — and until now, that meant starting from scratch. Every dismissed false positive would reappear, and you'd have to manually clear it again.
Version control fixes this. Upload a revised file, and Auto QC re-applies the dismiss states from your previous check automatically. Only genuinely new issues surface.
How it works
Open the action menu on any completed analysis and select “Upload New Version”. Auto QC runs a full quality check on the new file using the same settings as the previous version — no reconfiguration needed.
- 01Open the action menu on any completed analysis → "Upload New Version"
- 02Select the revised video file
- 03Auto QC analyses the new version with inherited settings
- 04Dismissed errors are re-applied automatically — only genuinely new issues appear
Automatic error state inheritance
The tricky part of version QC isn't detecting errors — it's deciding which ones were intentional. A black frame you dismissed because it's a deliberate fade-to-black should stay dismissed. An audio clip you ignored because it's an approved effect shouldn't pop up again.
Auto QC uses a three-layer matching system to handle this:
Aspect ratio, Loudness
Type match — no timecode needed
Black frames, Freeze frames, Audio clips, Blur...
Audio alignment → timecode translation → ±0.5s window
Typos, Safe zone violations
Text similarity + aligned timecode
Timeline alignment across edits
The hard case is when the edit shifts the timeline. Cut two seconds from the head, and every timestamp in the QC report is now two seconds off. A straightforward time-based match would miss everything.
Auto QC solves this with audio cross-correlation. We extract a 50Hz RMS energy envelope from both versions and run a sliding-window Pearson correlation to find the exact offset. The approach is codec-agnostic — it works after re-encoding because audio energy is preserved across codec changes, even when the raw PCM data looks completely different. Typical accuracy is within 0.02 seconds.
ALIGNMENT METHOD
- Extract 8kHz mono PCM from both versions using FFmpeg
- Compute 50Hz RMS energy envelope (20ms windows)
- Z-score normalise, then run sliding-window Pearson correlation
- Search forward and backward offsets up to 60 seconds
- Translate V1 timecodes into the V2 timeline before error matching
When alignment confidence is high enough (≥ 0.65), all V1 timecodes get translated before matching. When it's lower — the audio changed significantly, or the video has no audio track — Auto QC falls back to inheriting global errors only and shows a manual review prompt. It won't silently inherit the wrong thing.
What carries over, what doesn't
Dismissed errors are re-applied to the new version. Errors marked as “corrected” are not — if you fixed something in the edit, you want to confirm it's actually gone in V2, not silently inherit a resolved state.
Analysis settings also carry over: detector toggles, platform compliance preset, PlanD configuration. You don't need to reconfigure anything unless the new version has genuinely different requirements.
The dashboard
The dashboard shows the latest version of each video. Previous versions are accessible from the analysis result page via the version history panel. A small version badge (V1, V2, V3...) on each dashboard card makes clear which revision you're looking at.
Frequently asked questions
What happens to dismissed errors when I upload a new version?
Dismissed errors are automatically re-applied using a three-layer matching system. Global errors (aspect ratio, loudness) match by type. Timeline errors match via audio cross-correlation alignment with ±0.5s tolerance. Subtitle errors match by text similarity plus aligned timecode.
Does version control work if I re-encode the video before uploading?
Yes. Auto QC uses audio cross-correlation rather than raw audio data for alignment — audio energy is preserved across codec changes. The offset is found correctly even after re-encoding. Typical accuracy is within 0.02 seconds.
What if my edit shifted the timeline by more than 60 seconds?
Auto QC searches offsets up to 60 seconds in either direction. If the shift exceeds this or alignment confidence is low, it falls back to inheriting global errors only and shows a manual review prompt — it will not silently apply incorrect timecode translations.
Which settings carry over to a new version automatically?
All analysis settings: detector toggles, platform compliance preset, and PlanD configuration. No reconfiguration is needed unless the new version has different requirements.
Version control is available on all plans, starting today. See all Auto QC features →
Also in the AI detection suite: Logo & Brand Detection — automatically flag uncleared trademarks before delivery