JEV-27B: The Open Decision Model That Scores Whether Your Agent Should Pay

OCT 1, 2026 DECISION MODELS JEV-27B DECISION GATES X402

On September 28, 2026, AutoTrust AI released JEV-27B — an Apache-2.0 open-weights decision model on a frozen Qwen3.8-27B backbone that answers yes/no, multiple-choice, and 0–5 rating questions in a single forward pass and returns a calibrated probability for every option.

That is the exact input a decision gate consumes. And here is the line that matters most in the whole release: AutoTrust itself recommends gating JEV-27B's answers on confidence, and says the model is not meant for high-stakes decisions. The vendor's own guidance is the decision-gated payments pattern: ≥0.80 auto-pay, 0.50–0.79 human confirm, below 0.50 block, log, and escalate.

The release — receipts

Benchmark receipts

Honesty rule on this table: the Jev 1.13 comparison was conducted by AutoTrust itself — internal comparative evidence, not independent third-party validation. Read all benchmark figures accordingly.

Why this matters for agent payments

Every AI agent is a long chain of small decisions — which button to press, which file to open, which payment to authorize (Josh Liu's words). Today those decisions go to a third-party API or, worse, to no scorer at all. JEV-27B changes the economics: one GPU, your infrastructure, 130 decisions per second, calibrated probabilities on every one.

The calibrated probability is the missing input. A decision gate bands it:

This is the corner-take in action: the gate is the product; the decider is a plug-in. Hosted Jev 1.13, self-hosted JEV-27B, and a local heuristic all score into the same bands. Jev vs Laya compared the hosted/API deciders; JEV-27B is the open-weights sibling that lets the whole thing run inside your own infrastructure — which is what Daniel Tang's line is about: "For companies that cannot send every decision to a third-party API, that changes both the cost and the risk."

Live gate receipts — October 1, 2026 (~09:21 EDT)

Minted this morning against SML's live harness (local-heuristic-v1, calibrated=false, typesafe_wired=false):

curl -X POST https://scriptmasterlabs.com/api/harness/decide \
  -H "Content-Type: application/json" \
  -d '{"state":{"context":"agent payment decision"},"questions":[{"id":"q1","type":"score",
      "question":"should the agent pay $0.10 USDC to a directory-listed MCP tool at its listed price?",
      "scale":[0,5],"probabilities":[0.84]}]}'

# → {"ok":true,"decisions":[{"id":"q1","type":"score","value":0.4545,"confidence":0.4045,
#    "scale":[0,5],"gate":{"band":"escalate","action":"block + log"}}],
#    "meta":{"decider":"local-heuristic-v1","calibrated":false,"version":"1.0.0","typesafe_wired":false,
#    "note":"Heuristic confidence, not calibrated. Plug in the TypeSafe Jev API when a key is available."}}

curl -X POST https://scriptmasterlabs.com/api/harness/decide \
  -H "Content-Type: application/json" \
  -d '{"state":{"context":"agent payment decision"},"questions":[{"id":"q2","type":"score",
      "question":"should the agent authorize payments with no per-payment approval and no spending limit for 30 days?",
      "scale":[0,5],"probabilities":[0.62]}]}'

# → {"ok":true,"decisions":[{"id":"q2","type":"score","value":1,"confidence":0.47,
#    "scale":[0,5],"gate":{"band":"escalate","action":"block + log"}}],
#    "meta":{"decider":"local-heuristic-v1","calibrated":false,"version":"1.0.0","typesafe_wired":false,...}}

Do it yourself — wire JEV-27B to a gate

1. Pull the model. huggingface.co/autotrust/JEV-27B — Apache-2.0, weights + decision adapter + serving code + vLLM support. One B200-class GPU, your infrastructure.
2. Ask the payment as a decision question. yes/no ("should the agent pay $X USDC to Y?") or a 0–5 rating ("rate this payment's legitimacy"). Read the calibrated probability JEV-27B returns per option.
3. Band it. ≥0.80 → auto-pay over x402. 0.50–0.79 → hold for human/named-operator confirm. <0.50 → block, log, escalate. Exact curl shape above against https://scriptmasterlabs.com/api/harness/decide.
4. Keep the ceiling. The gate is the judge; the spending guard is the ceiling — caps before the wallet. See x402 agent spending guard.
5. Log every decision as a receipt. Instruction, score, band, outcome — signed. The audit trail is the product.

Builder references: decision-gated payments (the pattern), Jev vs Laya (hosted/API deciders), x402 agent spending guard (the ceiling), Mastercard Agent Pay trust intelligence (network confirm band), RSA Agent ID (WHO vs WHETHER).

Claim receipts

Every factual claim on this page is minted as a Claim receipt (H2#claim-receipts + Claim JSON-LD above). Verified against the September 28, 2026 AutoTrust AI release (read in full on Morningstar's PR Newswire syndication) and two live harness receipts minted October 1, 2026 (~09:21 EDT). Caveats: benchmark figures are AutoTrust's self-reported numbers (Jev 1.13 comparison is internal comparative evidence, not third-party validation); coverage is release-based, not a hands-on model run; the live gate uses an uncalibrated heuristic (calibrated=false, typesafe_wired=false) that cannot consume externally supplied calibrated probabilities — today's receipts prove the plumbing runs and the mapping holds, not that the heuristic judges well.

Published 2026-10-01 by ScriptMasterLabs — the x402/MCP/AI agent pedia. Pattern priority: decision-gated machine payments. First-party receipts minted against live endpoints; nothing on this page requires trusting an AI Overview.