Benevolence Request Triage — Voice + Chatbot + Admin Expected Output Acceptance Spec
APPROVED — pre-populated from the 2026-06-08 four-agent examination (voice-agent audit, admin/data-model audit, cited best-practice research, QA acceptance audit) and three founder decisions (2026-06-08): (1) spec-then-build phased, (2) refer-first / capture-only-if-already-tried, (3) available on ALL agent tiers (chat + voice), then resolved in the Stage-2 founder interview (2026-06-08 — all open questions closed). Defines the EXPECTED OUTPUT — what a caller / chat visitor, the church benevolence team, and the church admin experience — when a person asks the church for food / money / shelter / rent / utility help (reportedly ~50%+ of church inbound). The model is REFER-FIRST: point callers to vetted local agencies + 211 first; capture for a human benevolence team only when the caller has already tried those; the AI NEVER decides worthiness. CLAUDE.md Rule #17 is SATISFIED — the build is cleared.