I tried this on a around 400 frames of the Park City UT video, but running it a couple of times on the same section gave different results. For example the first with Vega identified, and second with Achird visible. But, perhaps that video is too blurry to be useful?

I'd closed it since but just tried it again. https://www.metabunk.org/sitrec/?custom=18302/StarTrackerTest_1/20260804_203008.js
From frames 1235 to 1413 (20.23s-23.14s) - (I was curious what it would make of the five faint stars in a chevron shape).
It now gives the same result each time.

but then when I adjust the end point to 31s and run it again (671 frames) it thinks the camera is looking at the different SE sky.
But again, without making any changes and running it again it gives the same result. So, I haven't been able to replicate the two different results from my previous post without making a change to the number of frames.