Story

In brief

We have lots of benchmarks for new frontier LLMs (SWE benchmarks etc) to make them score on the "best code". In large codebase, the best code is the one that matches the codebase own voice and conventions, because tha...