Parsimony

Parsimony measures how much implementation code AI coding agents add or remove while attempting benchmark tasks. Its author compares syntax units rather than lines, so short formatting and compressed code do not automatically improve a score. The public site shows a leaderboard and the GitHub repository supplies source code, method notes and published measurements. The repository predates October 13; the current attention comes from its English Show HN submission. Parsimony relies on benchmark-published correctness results instead of rerunning every test, and a smaller patch is not necessarily a clearer or better fix.