Quality Pipelines for Generative AI
summary
Case study: STA Scala Center work on the Scala standard library uses Scala programs to insert TODO markers, a long-running orchestration script that runs a writer/reviewer/adjudicator pipeline (Opus, Sonnet, Codex/Claude Code), and deterministic programs to reformat Scaladoc and convert WikiDoc→Markdown. The system produces one commit per milestone and one PR per run, enforces a green gate test run before review, generated 21 focused PRs for param and style fixes, and found 350 issues in a book preflight pass.
AI-written summary · about
