Agent Trust Gate · Development evidence · 20 October 2026
Controlled local demonstration · no real purchasing or payments
Genuine Microsoft reviewer identity. Separate authority for a synthetic £4,000 purchase.
ATG has completed one controlled, non-production workflow combining genuine Microsoft Entra reviewer authentication with a separately approved £4,000 synthetic purchase.
Identity admission granted zero purchasing authority. The operator separately confirmed approval of the exact synthetic action and then confirmed execution. Retained local evidence shows one execution, linked decision and execution receipts, consumed one-use authority and no duplicate or UNKNOWN outcome.
No real purchase, payment or external purchasing API was used. Microsoft authenticated identity; it did not approve spending. Action authority used public synthetic fixture keys and fixed action time. This is a local companion demonstration, not production qualification or independent attestation.
What was verified
- All identity admission authority flags were false.
- GatePass signature and reviewer, exact-request and business-state bindings matched.
- One journal attempt and destination order; lifecycle EXECUTED and exposure COMMITTED.
- Distinct decision and execution receipts with matching links and integrity digests.
Changed details, identity substitution, replay, expiry and uncertain outcomes are covered by offline cases. No live replay or changed-detail attempt was made in the successful run.
Separate components, separate claims
The earlier offline buyer demonstration still uses mock Microsoft identity. The published local.7 plugin and the separate operational record-service candidate retain their existing capabilities and qualification status. This result neither repackages Microsoft login into local.7 nor qualifies the operational candidate's own sign-in, MFA, TLS or deployment.
Proposed £1,500 evaluation
Evaluate one synthetic purchasing workflow and up to five agreed acceptance questions. Receive a guided walkthrough up to 90 minutes, sanitised evidence review, written findings with integration gaps and a GO / conditional GO / NO-GO recommendation, a handover up to 30 minutes and one factual-correction pass.
The default buyer session is offline and synthetic, with mock identity and no buyer credentials or system access. Prior genuine sign-in evidence can be discussed through this sanitised summary. A new live sign-in or buyer integration needs separate scope and approval. Fee subject to scope, availability and written agreement; tax, payment terms, dates and acceptance arrangements remain to be agreed.
Next step: nominate one workflow, a workflow owner and a technical reviewer, and agree up to five acceptance questions.
View the evaluation presentation · Sanitised validation facts · Failure history and limitations · Discuss one workflow
Limits remain visible
One local run is not independent audit evidence, Microsoft endorsement, MFA assurance, regulatory certification or production readiness. Historical Windows EPERM failures remain unattributed. Same-host records are unencrypted; same-user/admin compromise is outside the demonstrated boundary. Fixed receipt dates are synthetic, not proof of wall-clock execution time. No customer names, real identities or raw authentication evidence are distributed.