Cross-Model LLM Code Review: Should you use Claude to review Codex or vice versa

I love this experiment. I think that the reviewer procedure can be improved with CLAR (e.g., the Impasse skill) or something similar, so I hope they try other approaches to their experiment.

From the original paper abstract:

“Claude review raises Codex drafts from 71.6% to 89.7%; Codex self review raises them to 84.5%. The reverse direction does not pay off: Codex reviewing Claude drafts drops the pass rate from 91.4% to 82.8%, and Claude self review leaves the 91.4% baseline unchanged.”

The highest pass rate was to use just Claude, at 91.4%. Claude reviewing Claude did not improve the pass rate.