Coding Models Are Doing Too Much

The author programmatically corrupts 400 BigCodeBench problems with single-point bugs to test whether coding models over-edit code during fixes. The post defines the minimal fix as exactly reversing the corruption and measures excess changes with token-level Python Levenshtein distance. The provided body does not disclose final results, model rankings, or training gains.

Why it matters: Strong HKR-K from a concrete 400-task bug-injection eval and a clear minimal-patch metric. HKR-R also lands because over-editing is a daily pain point for Copilot/Cursor/Claude Code users, but the excerpt omits results, model rankings, and effect sizes, so this sits near the low