Brief coordinator runbook

Authoritative source, reproduced unchanged. Exact Markdown · Agent copy with resolved links. These instructions govern research runs, not checkpoint interviews. They keep the run's own words: a room is a community, a mark an artwork, the analyst the aggregator, the case lead the community coder and the global model the population model.

1. Inspect python3 _tools/brief_queue.py list --limit 10. Work only on queued ranks in that order.
2. Keep at most six active writers. When one writer completes, handle that single rank immediately, then refill its slot. Do not wait for a batch.
3. Writers are always fresh agents: gpt-5.6-sol, high reasoning, fork_context:false. Read BRIEF-WRITER-PROMPT.md afresh and state the exact rooms/<subreddit>/brief directory in the launch message.
4. For writers: prepare, spawn, start. At completion run finish --rank N --message ..., inspect returned JSON, then close the accepted worker.
5. A finish process exit of zero can still return status: retry. Retry only that rank with a fresh Sol writer. For an interrupted worker use recover --rank N --error ..., close it, then start a fresh writer.
6. For a work: harness rank with zero pre-week items, run prepare and finish without a writer, then refill the resulting slot.
7. Never use a resumed production writer. Never process event coding or excluded prior-run rooms. Do not use the legacy sync script while this moved yard is canonical.
8. At completion, confirm 395 verified, no running or queued ranks, run the brief validation and reader tests with PYTHONPATH=_tools python3 -m unittest _tools/test_brief_validation.py _tools/test_brief_reader.py, then pause the scheduler.