Held in 10 of 28 repos
This page shows whether PRs that breached a rule really needed a fix more often, on PRs the rules had never seen.
On PRs the rules had never seen, PRs that breached a rule needed a fix 1.7x as often (18.3% vs 10.7% of PRs that change code, 28 repos).
Repos where it held
Code PRs the rules had never seen, per repo. Needed a fix: a later PR had to fix it.
| Repo | Breached a rule: needed a fix | Breached no rule: needed a fix | Verdict | Over time |
|---|---|---|---|---|
| honojs/hono | 10 of 48 (21%) | 1 of 28 (4%) | Held | Held in 4 of 5 periods |
| langchain-ai/langchainjs | 6 of 17 (35%) | 5 of 54 (9%) | Held, early | Held in 3 of 4 periods |
| medusajs/medusa | 7 of 76 (9%) | 3 of 75 (4%) | Held, early | Held in 1 of 2 periods |
| microsoft/playwright | 17 of 67 (25%) | 13 of 114 (11%) | Held | Held in 3 of 3 periods |
| nuxt/nuxt | 33 of 125 (26%) | 9 of 91 (10%) | Held | First period tested |
| PrefectHQ/prefect | 9 of 112 (8%) | 2 of 55 (4%) | Held, early | Held in 2 of 2 periods |
| remix-run/remix | 3 of 26 (12%) | 0 of 72 (0%) | Held, early | Held in 2 of 4 periods |
| sveltejs/kit | 4 of 11 (36%) | 0 of 17 (0%) | Held, early | Held in 4 of 4 periods |
| vitejs/vite | 13 of 75 (17%) | 1 of 18 (6%) | Held | Held in 4 of 4 periods |
| vitest-dev/vitest | 14 of 73 (19%) | 0 of 29 (0%) | Held | Held in 4 of 4 periods |
| All 28 repos tested | 217 of 1,188 (18.3%) | 195 of 1,821 (10.7%) | 1.7x as often | |
Pooled numbers include all 28 repos tested, including the ones where it did not hold. "Over time" shows how often it held in earlier periods too.
Method
Rules are learned from what happened after past merges, then tested on later PRs they hadn't seen, with the same checks the GitHub Action runs.
The "Recent PRs that needed a fix" table on each report uses rules learned from that same history, so it shows how well they describe it.