tbaederr wrote: > Maybe we should change the existing evaluator to be consistent? Otherwise LGTM I'm not 100% sure what output is better, but we can change that later. This change is more about correctness. https://github.com/llvm/llvm-project/pull/67131