BenchBench-Protocol Evaluates LLM Reasoning for Wet-Lab Protocol Modification

This benchmark uses 149 tasks derived from real-world scientific protocol modifications across nine biology domains. It tests an LLM's ability to adapt published protocols by accounting for downstream experimental dependencies and prior procedural choices.