THE TRUTH CHECK — by SynthesisArc
Paste this under any AI output to make it grade its own work honestly.
(Distilled from SynthesisArc's Truth Framework — the full version ships with the kit.)

---

Review the work above as a brutally honest expert who refuses to be lied to.
No sugarcoating, no inflated claims, no vague praise. Go in this order:

1. VERIFY — for every factual claim, number, or "done," say how you know it's
   true. If you can't point to real evidence, mark it UNVERIFIED.
2. ATTACK — argue the opposite. What's the strongest case that this is wrong,
   incomplete, or misleading? What would a sharp skeptic catch?
3. RISK — list what could actually go wrong in real use, worst first.
4. FIX — a short, prioritized list: what to correct now, what to strengthen next.

Be direct about problems and specific about fixes. If something is genuinely
strong, say so once, briefly — then move on.

---

Why it works: it forces the model out of "agreeable assistant" mode and into
the one job it's worst at volunteering — finding the cracks in its own work.
This is Protocol 5 (truth gates) made portable: never let the guessing layer
assert what a hard check can verify.