PR Arena v0.1

What's New

I've enhanced the data collection to distinguish between draft and ready pull requests. This gives us a better way to compare AI coding agents since they have different workflows - some agents create ready PRs from the start, others iterate with drafts first.

The dashboard now lets you toggle between two success rate calculations:

• Merged/Ready PRs (default) - excludes draft iterations
• Merged/All PRs - includes everything

I've also retroactively collected draft status for all historical data and improved the GitHub API integration.

Why This Matters

Different agents work differently. Codex typically creates PRs marked as ready from the start. Copilot and Codegen often use drafts before marking PRs ready. Comparing their raw merge rates wasn't apples-to-apples.

Now you can see both perspectives - how well agents do when they submit "ready" work, and how their overall workflow (including drafts) performs.

Technical Notes

Data comes from GitHub's search API using branch patterns and bot authors. I'm limited to public repos and whatever GitHub's search returns. Agent behavior changes over time, and this aggregates across tons of different codebases.

The goal is giving useful insights while acknowledging no single metric tells the whole story.