# I lost to two AI models on my own bug fix

> Source: <https://www.catchkill9.dev/posts/sync-by-issue-number.html>
> Published: 2026-11-20 00:00:00+00:00

**Recommendation:** When fixing a number-keyed matching bug, audit every code path that resolves the same key — pull and push — before submitting, since the graded pipeline tested both and the human fix covered only one.
More in the playbook "Working with AI coding agents": <https://wpnews.pro/playbooks/coding-agents>

# I lost to two AI models on my own bug fix

A pipeline graded my beads PR against glm and opus. I fixed pull and never checked push.

I opened a PR on an open-source project and lost to two AI models.

Not rejected. The maintainers run every fix through a pipeline that rebuilds it with other models and scores all versions side by side. Mine: partial on four acceptance criteria, failed the fifth. glm and opus passed all five.

The fifth was push. I had fixed bd github pull, which matched issues by number only, so issue 42 from a second repo overwrote the bead linked to the first repo's 42. I never checked that push had the same bug going the other way. Both models did.

Ten days earlier, a fork maintainer had already taken my fix, all five test cases included, and thanked me by name.

The merged PR is built from the models' code and says "supersedes #7794, with credit". It shipped in a hotfix the next day.

My name is in that PR. My code isn't.
