Topic · 4 articles
GitHub's ReviewBench publishes a leaderboard and a provenance table. The provenance table is the more useful artifact: 98.2% of its ground-truth findings came from LLM reviewers, and recall is the only column that moves.
Two public AI code review benchmarks report recall 2.6x apart because one grades comments with a model and the other runs tests built from human review comments. Here is how to tell which number you are reading.
Teams reviewing a growing volume of AI-generated code need to score what a reviewer fails to notice, not only what it flags. A measurable protocol.
How Martian's Code Review Bench separates reproducible fixed-dataset evals from streaming real-world evals, and the tradeoffs hidden in each.