standup
Standup — 2026-06-24
Six PRs merged yesterday, all on one thread: the global-brain inbound ingest spine.
Daily standup for 2026-06-24. What shipped:
Six PRs merged yesterday, all on one thread: the global-brain inbound ingest spine.
ADR-0026 landed in #175. The decision: one shared staging table (global_brain_candidates), one review→embed-on-approve path, and one write-path guard for the sensitive-vertical exclusion — instead of three divergent pipelines for playbooks, research, and cross-pollination. Two authorization tiers: staff-authored playbooks write directly (PR-as-gate) with mandatory in-row provenance; machine- and customer-derived content must stage and pass human review.
Phase A shipped in #174: scripts/ingest_playbooks.py reads playbooks/*.md, embeds via Voyage voyage-3-lite, and upserts into brain_entries with tenant_id="global". Only status: approved files publish; drafts are skipped. Phase B Slice 1 followed in #179: the global_brain_candidates staging table, an OIDC-gated POST /internal/brain/candidates endpoint, consent-version enforcement, and idempotent upsert on (source_type, study_id, run_id). 2,097 tests pass; migration verified reversible.
Three nits from the cross-team review were fixed in #177: --dry-run no longer spends Voyage quota, missing ingest_commit provenance now logs a loud warning, and a bad --dir argument raises a clean error instead of silently misbehaving via glob.
ONBOARDING.md was added in #176 and corrected in #178 — covers the four surfaces, CI invariants, agent review gates, and the ADR sign-off flow.
Notable. The --dry-run Voyage spend was a real footgun: a preview run would have silently burned embed quota with no writes to show for it. Caught in review, fixed before anyone noticed the bill.
Window: 2026-06-23T13:00
→ 2026-06-24T13:00.824Z · Sources: memberintel @ 605d6e4, memberpress-intel @ 8481377