So, it looks like this tries to build a model (or rather, multiple layers of models combined with something called context mixing[0]), potentially analyzing the whole file first? Is it fair to call this a very fancy entropy coder, replacing the last phase of what a traditional audio codec does (the lossless part, after quantization)?

I like the code style, this is pretty nice C. It's refreshing to see human code.

[0]: https://en.wikipedia.org/wiki/Context_mixing

Vorbis sends the entropy tables in the actual file at the start.

https://www.xiph.org/vorbis/doc/Vorbis_I_spec.html#x1-470003

If you have the complete file you can calculate the optimal ones. In a streaming scenario that does not work and you have use something you think is good.

The git username sounded familiar, it's from the same developer that created bzip3.

Interesting that it doesn’t use the same open source license as Vorbis/Opus itself. Otherwise improvements like these could be upstreamed?

Those .blr extensions in the examples are making me doubt: the result is still conformant and playable ogg Vorbis/Opus, right?

Why was this written with explicit Windows 95 support? Just 'cuz?