Sampling at fixed timestamps works for this clip, but if a short has burned in captions, I'd take the sample times from the subtitle file instead. Grab one frame at the start of every cue in the .srt and the sheet shows every caption against your bottom 20% zone, which is where captions and the platform's own caption text collide most.

For software UI shorts the letterbox layout gets rough. Fitting a 1920 wide app window into 1080 pixels shrinks 14px body text to about 8px. What reads better is cropping to the one panel in use and scaling it up to fill the safe area, then running your check on that version.

The CI idea is the part I'd copy first. A QA sheet saved next to every export costs nothing and catches the end card mistake every time.