{"slug": "i-lost-to-two-ai-models-on-my-own-bug-fix", "title": "I lost to two AI models on my own bug fix", "summary": "A developer's pull request to an open-source project was superseded after a maintainer pipeline rebuilt the fix with two AI models, glm and opus, which passed all five acceptance criteria while the human submission passed four and failed the fifth, covering the push path. The developer 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, but never checked whether push carried the same bug. The merged PR, built from the models' code and labeled \"supersedes #7794, with credit,\" shipped in a hotfix the next day, ten days after a fork maintainer had already taken the original fix with all five test cases.", "body_md": "# I lost to two AI models on my own bug fix\n\nA pipeline graded my beads PR against glm and opus. I fixed pull and never checked push.\n\nI opened a PR on an open-source project and lost to two AI models.\n\nNot 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.\n\nThe 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.\n\nTen days earlier, a fork maintainer had already taken my fix, all five test cases included, and thanked me by name.\n\nThe merged PR is built from the models' code and says \"supersedes #7794, with credit\". It shipped in a hotfix the next day.\n\nMy name is in that PR. My code isn't.", "url": "https://wpnews.pro/news/i-lost-to-two-ai-models-on-my-own-bug-fix", "canonical_source": "https://www.catchkill9.dev/posts/sync-by-issue-number.html", "published_at": "2026-11-20 00:00:00+00:00", "updated_at": "2026-11-20 12:41:47.664421+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools"], "entities": ["glm", "opus", "bd github"], "also_reported_by": [], "takeaway": "A fix that matches issues by number only must be checked on both pull and push paths, since the same collision that overwrites a bead on pull can overwrite it on push.", "recommendation": {"advice": "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.", "playbook": "coding-agents", "playbook_url": "https://wpnews.pro/playbooks/coding-agents"}, "alternates": {"html": "https://wpnews.pro/news/i-lost-to-two-ai-models-on-my-own-bug-fix", "markdown": "https://wpnews.pro/news/i-lost-to-two-ai-models-on-my-own-bug-fix.md", "text": "https://wpnews.pro/news/i-lost-to-two-ai-models-on-my-own-bug-fix.txt", "jsonld": "https://wpnews.pro/news/i-lost-to-two-ai-models-on-my-own-bug-fix.jsonld"}}