Standup — 2026-07-15

▶ Watch the 46s summary — Daily standup — Standup 2026-07-15

Daily standup for 2026-07-15. What shipped:

Production eval monitoring shipped. #434 hooks sampled quality scoring into llm/call.py’s existing post-response seam — 5% of CHAT, 20% of SITE_ANALYSIS — with code-assertion checks (anti-genericness denylist, citation validity, term-overlap groundedness proxy), two new tables, and cost-breaker-style alerting. No judge model, no third-party service. ADR-0035 and the ai-engineer build-vs-buy review landed in #431; Blair’s verbal go-ahead is recorded but the written ADR still needs formal sign-off.

Six more PRs merged on the admin console:

• #433 — admin shell now proves identity via GET /admin/whoami at load; a CF-allowlisted non-admin gets an honest “no admin access” screen instead of a fake SUPER ADMIN label followed by misleading session-expiry errors on every page (Ronald hit this exact flow on 2026-07-14)
• #432 — audit log gains offset-based pagination and a copyable raw-JSON modal
• #436 — Brain corpus filter sidebar collapses by default, auto-expands only groups with active selections; facet state persists in the URL as a shareable link
• #438 — subbrain Sites tab reskinned onto --mi-*, now shows per-site industry/archetype chips and a sync-failure badge; fixes billing period singularization (/months → /month), member count pluralization, and blank-line rows in the BIBLE editor
• #426 — email identity converted to Postgres citext; case-variant addresses now resolve to one account at the DB level across all auth paths
• #435 — pre-beta hardening: downgrade-cycling detection (free→pro→free ≥3x in 90 days writes a billing.cycling_detected audit row, visibility-only) and a guard against a non-str source_text in the eval-monitoring SITE_ANALYSIS branch

Also merged: admin overview dashboard #425, batch-7 playbooks #428 (30 entries), ingest-history mi-* reskin #430, and the Brain Content Editor design spec #437.

On the KB side, Seth registered the complete eight-cut redub meme ad series (#108–#115) and grouped the /video-review shorts into six collapsible series sections (#111).

Notable. ADR-0035 is in “Proposed” — the eval-monitoring system is built and running on staging, but Blair hasn’t read and approved the written artifact. It should move to Accepted before the system samples any real production customer traffic at V1 launch.

---

Window: 2026-07-14T13:00

→ 2026-07-15T13:36

.864Z · Sources: memberintel @ 4cb556c, memberpress-intel @ df768b9