Show HN: Mdarena — Benchmark your Claude.md against your own PRs

What it is

Mdarena is a new open-source project on GitHub that promises a practical way to benchmark Claude.md outputs against your real pull requests. It has been reported that the repo, posted as a Show HN thread, organizes model-generated Markdown and your repository’s diffs so you can compare suggestions, track regressions, and quantify how often the model’s edits would have passed or failed your code review. Simple idea. Useful, too.

Why this matters

Developers are using LLMs as co-pilots more than ever. But how do you measure whether those co-pilots help or hurt? Mdarena aims to cut through the noise by putting model suggestions next to the stuff you actually ship. Can an AI teammate be judged by the same metrics as a human? That question feels sudden and urgent — and that’s the emotional core here: relief for cautious teams, and accountability for teams riding the hype train.

How to try (and caveats)

The project lives on GitHub at the provided repository and appears to be plug-and-play for folks who want to run local benchmarks against their PR history. It has been reported that contributors are iterating quickly, but some claims about accuracy and coverage remain allegedly unverified until broader adoption. In short: promising tool, early stage. Worth a spin if you want to audit your AI suggestions instead of taking them on blind faith.

Sources: github.com/hudsongri, Hacker News