Ah, got it, thanks.
I don’t think that we have definitive proof that the video files are not being generated every time, but most evidence points to that. (There might be some metadata in the mp4 that could help resolve this more definitely)
Considering the complexity of doing on the fly video splicing (even if the results are cached) vs the ability for big studios like WB to do the splicing offline in a controlled environment that they’re used to, it’s probably safe-ish to assume everything was generated in advance.
It also definitely helps with QA - you can be more sure that the splicing worked if you can test some/any/all files before serving them via CloudFront vs relying on a complex-enough piece of tech.
A breakdown of the tech behind this from WB would be interesting to read for sure!
Cheers