A voice agent isn’t launched once.It’s kept honest — for its whole life.

Lints a voice prompt against four safety blocks as hard gates — disclosure, no premature confirmation, barge-in, escalation — so a prompt that reads well but has no guardrails can't ship.

Content-hashes what's running in each live agent against your approved versions, catching the console edit that dropped a guardrail in production — a shadow prompt no one signed off on.

Scores each release against your own baseline on voice behavior a text diff can't see — barge-in, latency, dead-air, escalation — so a change that makes the agent talk over callers gets caught.

Grades real call transcripts and flags the ones the dashboard marks 'resolved' that a human would call mishandled — the hallucinated confirmation, the failed handoff.

Reconciles the bookings the agent claimed on calls against your calendar, catching the phantom appointment — a confirmation the caller heard that never made it to the book.

Finds the suppressed numbers still sitting in your dialing lists, normalizing phone formats so a do-not-call contact can't hide behind a different spelling of the same number.

Resolves every queued call into the called party's local time and holds the ones that would fire outside the calling window — the timezone flip an outbound queue silently gets wrong, since the queue is built in your zone but the window is theirs.

Catches PII spoken aloud — a card number read digit by digit — and left in the transcript, then re-scans its own output and flags anything that survived. Document scanners miss this.

Assembles the warm-transfer briefing a human needs on escalation and grades whether it's complete, so no caller restarts from zero. Drafts and grades only — never transfers the call.

Turns your retention policy into a dated purge schedule and flags what you're over-retaining — while never scheduling a class under legal hold. Deletes nothing; schedules only.