Counting bugs is the hard part of comparing AI review tools

Model benchmarks for code review usually report the same two numbers: how many bugs the tool found,...