Bounded · local-first · evidence-led

Evaluate one exact-action workflow.

Agent Trust Gate™ can be evaluated against one buyer-selected AI-agent action family using synthetic or explicitly approved non-production data. ATG checks declared authority, exact-action integrity, policy conditions and lifecycle state; the buyer retains every real execution boundary.

ATG does not decide whether an action is commercially wise, beneficial or correct.

One workflowA fixed exact-action family and agreed success criteria.
Local-firstNo production system or external execution connection.
Buyer-controlledThe buyer owns authority evidence and every real action boundary.
Evidence-ledDeterministic outcomes and reproducible local artefacts.
Evaluation flow

Mandate, exact action, verdict, evidence.

The pilot reuses the same exact-action core demonstrated in M161 and evidenced in M162. It creates no parallel policy authority and no live execution path.

01

Bound the workflow

Agree one action family, synthetic inputs, policy limits and evaluation criteria.

02

Submit exact proposals

Exercise authorised, altered, review-band, replay and lifecycle cases.

03

Evaluate deterministically

ATG returns ACCEPT, REJECT or REFER; revocation remains an internal lifecycle control.

04

Review evidence

Inspect separate decision, lifecycle and synthetic execution evidence where applicable.

Fixed safety boundary

Evaluation does not become execution authority.

  • Only atg.evaluate_action is exposed over local MCP stdio.
  • Shadow Mode issues no GatePass and grants no authority.
  • Customer Trust Receipts are non-authorising and cannot act as GatePasses.
  • No live procurement, payment, settlement or customer communication occurs.
  • No production credentials, secrets or unnecessary personal data are accepted.
  • No automatic retry follows an uncertain external-effect state.
Execution remains separate

The buyer retains control of any real execution boundary.

The current public demonstration uses one frozen in-process synthetic adapter. Its acknowledgement is not a real order, payment or business outcome.

Evidence basis

Use verified synthetic observations, not promises.

Observed in the deterministic ten-case synthetic demonstration. These results are not customer, production, statistical, compliance, ROI or real-world risk-reduction evidence.

1/10 ACCEPT6/10 REJECT · 2/10 REFER · 1/10 SHADOW
18 demonstrated controls2 partial · 3 not demonstrated · 2 out of scope · 25 total
0/1 Shadow GatePassesShadow observation granted no execution authority.
0 automatic retriesAfter the uncertain crash/recovery state.
Commercial posture

Paid evaluation pilots starting from £1,500.

This is indicative and scope-dependent, not a quotation, checkout, guarantee or evidence of customer adoption. The evaluation is bounded, local-first, evidence-led and synthetic or non-production unless separately agreed and safely designed.

Contact Agent Trust Gate

Visible limitations

No production readiness, customer deployment or validation, regulatory approval, certification, guaranteed compliance or safety, proven ROI or loss reduction, statistical significance, distributed-system guarantee, real procurement/payment capability, business-outcome correctness or real external-effect reconciliation is claimed.