Vertical Caption QA Gate (QA)

Vertical Caption QA Gate (QA)

1. 01

   Reject any clip where word-level caption timing drifts more than 200ms from the spoken audio

   Scrub the first and last 5 seconds of each clip at 0.5x speed; drift compounds at clip boundaries where auto-cutters splice mid-sentence.

2. 02

   Confirm caption safe zones clear platform UI overlays on all three aspect ratios

   TikTok and Reels bury the bottom 15 to 20 percent under buttons and handles; check the burned-in render, not the editor preview.

3. 03

   Verify brand lexicon and product names against the client's approved spelling list before export

   Auto-transcription mangles proper nouns and acronyms consistently; a 12-term client glossary catches most of it in one pass.

4. 04

   Sample two clips per batch for narrative coherence, not just caption accuracy

   Automated moment detection optimizes for hook density and will happily cut a sentence in half if the pause reads as a beat.

5. 05

   Check that b-roll and auto-inserted visuals do not contradict the spoken claim

   Stock footage matched on keyword rather than meaning is the most common source of client-facing embarrassment in automated edits.

6. 06

   Log every manual correction by category so recurring failures get fixed at the template level

   If the same caption error appears in three consecutive batches, the fix belongs in the brand preset, not in the reviewer's queue.

7. 07

   Require a named human sign-off on the batch before any scheduled post goes live

   Schedulers inside these platforms publish on a timer; an unsigned batch that auto-posts is a retainer-level risk, not a minor slip.