The audit pipeline a 10–100 person agency actually needs.

Every AP invoice that lands in the agency's shared inbox — scanner, forwarded email, vendor portal — is captured into the same audit lane. Routing is by remit-to and cost-center, so a Boston print PO and a London retainer invoice don't fight for the same line. The audit fires the moment the message lands, not three days later when somebody sees it.

Retainer SOWs with open end-dates, milestone POs with stale unit prices, and one-off purchase orders against a new vendor all reconcile against the same canonical PO list. A line that matches cleanly continues straight through; a missing end-date, a short delivery, or an out-of-band total pauses the posting and surfaces on the morning digest with the rule, the reason, and the suggested next action.

A freelance illustrator onboarded outside the master, a remit-to change that looks like a shell company, a SaaS bill re-posting at a slightly drifted amount — the four most common agency vendor-stack failures all run against the same audit. Any one of them pauses posting, scores the rule, and lands on the digest the next morning instead of three months later at rec.

A clean invoice posts through the official API on a post-only OAuth token, carrying its confidence score, the matched PO, and the canonical cost-center. A thirty-day reversal window backs every post in case the morning digest returns a thumbs-down. The controller sees only what failed a rule, and what posted — not every line in between.