Jev vs Sonnet triage: the 9 questions and criteria A developer triaged 1,895 open GitHub issues across n8n, Supabase, Cal.com, Appwrite and Home Assistant using a fixed set of nine questions — covering issue kind, reproducibility, version reporting, expected behavior, actionability, severity, frustration, data integrity and workarounds — asked in a single call per issue. Comparing the "kind" classification against maintainer-applied labels on 767 issues, the Jev model scored 96.1% and Claude Sonnet 94.0%. | | The nine questions used to triage 1,895 open GitHub issues | | | n8n, Supabase, Cal.com, Appwrite, Home Assistant with Jev and Claude Sonnet. | | | | | | Both models got exactly this text: same questions, same one-line criteria. | | | All nine are asked in one call per issue. The issue goes in as: | | | REPOSITORY: