Indeed, I lost the history in a shuffle, but a similar use case broke in some Firefox update, and it's the exact reason behind this comment:

https://github.com/seligman/podcast_to_text/blob/master/sear...

In my case, loading the entire file is loading a tiny bit more data, so this fallback doesn't hurt, but it's still annoying, and broke any hope I had of doing something more clever with the dataset.

The actual regression might be with githup pages where firefox was sending the same ambiguous headers the whole time but something in github's stack started interpreting them differently.