> What I feel a bit annoyed by, and what I feel obviously LLM-run ablations like this fail to capture, is any kind of reflection around previous research or any kind of proof that this is the best you can do.

At least the former is easy: Claude is very happy to do a prior art search and pretty good at it, too.