M MemberIntel KB
Activity Decisions

standup

Standup — 2026-07-02

**ADR-0030 accepted.** [#247](https://github.com/caseproof/memberintel/pull/247) ships the v1a entry-level performance signal for the global brain: a nightly `e

Daily standup for 2026-07-02. What shipped:

ADR-0030 accepted. #247 ships the v1a entry-level performance signal for the global brain: a nightly entry_performance rollup (citation counts, 👍/👎, helpful rate — all-time and 30d) over global non-memory entries, surfaced in a new admin panel. Volume stops being the content rock; cited helpfulness does. Conditions open: Q3 numeric target and a named weekly pruning ritual (tracked under #11).

Playbook ingest got a three-PR fix run. Citation summaries in the Sources UI were pulling from ## Situation (when to use the play) instead of ## Move (what to do). Fixed in sequence:

  • #272 — rewrite _derive_summary to prefer ## Move, with fallback
  • #273 — strip raw markdown (bold markers, list dashes, trailing colons) from derived summaries; 37 live playbooks now produce clean text
  • #264 — purge [[wikilinks]] and **Related:** lines from served bodies (these were reaching the LLM prompt and the citation modal verbatim)

These fixes take effect on next staging re-ingest, not on merge.

Other merges: #269 adds a sensitive-vertical ingest backstop for benchmarks and reconciles faith/health vocab aliases — downgraded from stop-the-line to defense-in-depth after counsel review confirmed no faith/health benchmark row can be produced by the current pipeline. #265 wires auto-deploy of both frontends to Cloudflare Pages on merge to main (path-gated, Node 22 pinned). #262 type-guards citation metadata so null/non-string values can’t reach the client as "None". ADR-0029 closed. Playbook corpus now at 38 approved entries.

A counsel persona landed (#25 in the KB repo) to handle first-pass legal/privacy review while outside counsel is unresponsive. It can clear staging/alpha; prod go-live still requires Allen or Blair.

Notable. The helpful_rate denominator bug in the v1a rollup — dividing up-votes by feedback-bearing message count rather than up/(up+down) — meant the panel could display rates above 1.0 and was caught by ai-engineer review before merge. Fixed in-branch with a regression test.


Window: 2026-07-01T13:00

→ 2026-07-02T13:38
.723Z · Sources: memberintel @ bc4664a, memberpress-intel @ f224620

For: S Seth Shoultes