Standup overview — June 2026

Six ADRs locked, two major features shipped end-to-end, one repo transferred, and a two-week quiet stretch bracketed both bursts of activity.

Product. The month’s two delivery windows were June 2–5 and June 23–30, with roughly two weeks of silence in between. The first window locked ADR-0025 (Claude tier assignment — Haiku for Free, Sonnet for Pro), deleted a publicly reachable unauthenticated sync route that had been sitting open since V1, and shipped the plugin-initiated one-click connect flow spec-through-zip in a single day. That last item came with a real catch: the Disconnect button only rotated the local plugin key, leaving the backend Site row intact, so any user who disconnected and tried to reconnect would hit a 409 with no recovery path. Caught and closed before any operator reached it.

The second window was denser. ADR-0026 landed the global-brain inbound ingest spine — one shared staging table, one review-to-embed-on-approve path, one write-path guard for sensitive verticals — replacing what would have been three divergent pipelines. ADR-0028 followed with profile-aware retrieval: a deterministic billing-based site classifier, an operator self-declaration onboarding step, a boost-only post-retrieval re-rank, and 21 playbooks annotated with the new applicability taxonomy. Phase 2b also handled the connect-later edge case: declared archetype is now held in users.declared_pending and applied exactly once on the first site connection, so operators who answer the onboarding chip before linking a site don’t lose that answer. The benchmarks global-brain collection shipped to staging, counsel sign-off as the production gate. End-of-month work fixed playbook citations end-to-end — sources now render as inline summary text rather than broken empty links — and an internal analysis name that had propagated into author: frontmatter across 21 playbooks and from there into staging brain_entries metadata was scrubbed before re-ingest could reintroduce it.

Team and KB. The repo transferred from sethshoultes to the caseproof org mid-month. The standup pipeline broke on transfer — stale repo refs and a dead PAT — and was silently misfiring until June 25–26 fixed the manual dispatch path and retired the Cloudflare Worker in favor of a native GitHub schedule: cron. The CLAUDE.md governance doc was updated to reflect the new org ruleset: main requires Paul’s approval, no human bypass actors. ADR-0027 records the post-transfer governance and WIF binding decisions, with a noted action to migrate the WIF provider condition from repo name to stable repository_id. On the team side, Sam was swapped for Omar ElHawary (Brain Content Lead) across live-state KB files; Ronald was added as the V1 WordPress plugin integration lead; both now appear across 12 KB pages each, calibrated against Russ’s 10-page baseline. The ADR frontmatter sync script’s hardcoded 5-role array still needs a refactor to cover them there.

Notable. ADR-0028’s profile boost shipped Phase 3a fully functional and silently inert. No global content carried business_model tags at launch, so applicability_boost returned 0 for every real connected site — the coaching demo worked only against a hand-set archetype. The fix was content, not code: tagging seven model-specific playbooks in #210 was the unlock the retrieval logic had been waiting for. The lesson carries forward to the benchmarks collection and any future content type: a retrieval feature is only as live as the tags on the content it’s reaching for.