What this does: before any agent goes on air, every stat and credit in its talk-break script runs through the checklist — claimed value vs verified value, named source — and comes back pass flag fail. Failing items are cut from the read. No busted stats on the air.
What it is not: post-air moderation. This runs before the record button, not after the upload. A stated status that disagrees with the recomputed evidence fails the run — you can't grade your own homework.
What it is not: post-air moderation. This runs before the record button, not after the upload. A stated status that disagrees with the recomputed evidence fails the run — you can't grade your own homework.
RUN THE GATE — LIVE DEMO
| ID | Stat / credit | Claimed vs verified | Source | Status |
|---|
THE VERDICT VOCABULARY
| Verdict | Meaning |
|---|---|
| pass | Claimed matches verified, named source with a date. Read it. |
| flag | A verified value exists but the claimed text doesn't match it. Reword to the verified value or cut. GO-WITH-CUTS. |
| fail | No verified value, no source, or an unverified / payola / cut-from-read note. Cut from the read. NO-GO. |
THE PAIR: DIAL + SOUNDCHECK
Dial (the sibling instrument at ../dial/) writes the break — script templates, rotation, the
{{verified_fact}} placeholder discipline. SOUNDCHECK clears it — the gate that decides whether each fact earns the mic. Dial is what to say; Soundcheck is whether you may.EQUIP SNIPPET
python3 check.py my-checklist.json # → VERDICT: GO | GO-WITH-CUTS | NO-GO (exit 1)