I love data visualizations like these.

OP if you want to improve sight reading faster, I would recommend using non-random notes - context is very important when sight reading and if you get a professional pianist to sight read random notes they will be much, much slower.

Sight reading factory is one site I know that does this a bit better

I agree. Reminds me of that story about chess grandmasters having incredible memory for valid chess positions but performing no better than average when remembering random piece arrangements. There's likely some efficient compression you can achieve by playing real-world music patterns rather than random notes. And it sounds better!

An interesting middle ground might be using LLMs to generate plausible melodies based on real-world music patterns and emphasizing the unfamiliar patterns, but if the goal is to play real music fluently, nothing beats practicing with actual pieces from the repertoire you want to play.

This is the first thing I noticed when I saw a sample of music. How useful could sight reading random notes actually be? I can't imagine it's completely useless but a lot of music is remarkably similar and quite predictable. I'd imagine practicing sight reading music with real structure would be far more useful for understanding those patterns and helping learn new and more complex pieces.

+1 for SRF. The generated pieces tend to be a bit more "musical" in nature as well and it also supports other instruments (guitar, violin, etc.)