That's the question my company has been building against all year, and this week we published the measurements behind our answer — as a working paper, with the receipts attached, so anyone can check them or prove them wrong.
The answer is a number we call coherence. When an AI agent is about to act — release a payment, trade on a signal, assert a fact — independent evaluation paths vote for or against, and the gate computes:
coherence = |Σ w·d| / Σ w
—the weighted strength of agreement, compared against a threshold stated in advance. The rule that makes the number honest: the paths must be genuinely independent. Syndicated copies of the same story count once. A feed's own internal score never counts as a vote for itself. A path that can't answer abstains instead of guessing. Agreement among echoes is not agreement.
Three live measurements, each on its own instrument:
- Payments. On Arc mainnet (Oct 25, 2026), the gate scored a pending payment at 0.8125 against a 0.6 threshold — pass — and 0.01 USDC of a 0.25 authorization was captured while the rest was voided back, verified on-chain.
- Trades. PRO-LO v3.1 runs four independent paths under the same formula; its recorded live read is 0.9047. (Honest note: its jury layer still runs in shadow mode — the gate informs, it doesn't execute yet.)
- Facts. ClaimGate, our fact-checker for agents, returns SUPPORTED only at coherence ≥ 0.80 across at least two independent source families, with every supporting quote machine-verified against the fetched text. In its verification run it scored a true claim 0.8571 — and returned UNVERIFIABLE for a false one.
Full disclosure, because it's the whole point: these are my own measurements on my own systems, and the proof payments were me paying my own services — self-proof, labeled as such in the paper. There's no third-party benchmark here, and the paper's limitations section says exactly what hasn't been done yet, including independent reproduction and juror calibration. We published anyway — method, numbers, receipt IDs, and the dataset behind our 2026-11-03 census of the x402 ecosystem — because a number nobody can check is just marketing.
The working paper and dataset are on Zenodo: https://zenodo.org/records/24083889 (DOI: 10.5281/zenodo.24083889)
The gates are live, as paid agent services. ClaimGate fact-checks for agents at https://scriptmasterlabs.com/claimgate — and the same jury design decides in Verdict Protocol at $0.25 per decision, settled in USDC (https://scriptmasterlabs.com/sml-products). If you work on agent payments or agent trust, I'd rather hear where the method breaks than hear that it's interesting.
— Timothy Walton, Script Master Labs LLC (Service-Disabled Veteran-Owned Small Business)
FOLLOW
X: @TimmyCrypto78 — https://x.com/TimmyCrypto78
Instagram: @scriptmasterlabs — https://www.instagram.com/scriptmasterlabs
Threads: @scriptmasterlabs — https://www.threads.net/@scriptmasterlabs
Telegram: @x402sml — https://t.me/x402sml
LinkedIn: https://www.linkedin.com/in/timothy-walton-23224052
MY LINKS
🛒 All SML products: https://scriptmasterlabs.com/sml-products
Website: https://scriptmasterlabs.com
Books: https://scriptmasterlabs.com/sml-products#books
Shop: https://shop.scriptmasterlabs.com
Gumroad: https://timwalton.gumroad.com
Whop: https://whop.com/timmy741
Ko-fi: https://ko-fi.com/scriptmaster
Discord: https://discord.gg/jGWDutyVAR
Verdict Protocol: https://verdict.scriptmasterlabs.com
Free VA tools: https://va-ratings.org
Top comments (0)