Hmm this wasn't quite what I expected.

I thought that this might be some sort of highlight based on complexity/risk/importance. That immediately struck me as a "why didn't I think of that?!" moment when actually turns out they didn't either.

So I can imagine a scenario where you have a LLM agent's diff to review while iterating in an IDE. The diff would highlighted according to importance/complexity for human review. Trivia like getters and setters would be grey. Highly complex or complicated or risky code would be highlighted in some high-contrast bold colour (perhaps with a luminance scale depending on how critical it was). Then when you are reviewing the diff from the agent you can instantly see and focus on the most important bits that need the most attention.

I'd find that really useful.