Your dispatch layer executes anything shaped like a tool call — this service proves, per execution, that the call, the approval, and the control object were actually issued by the issuer they claim, for those exact arguments. Issuer signing at the source: the model-turn issuer, the human-approval service, and framework control issuers each sign a canonical digest of (kind, name, exact arguments, session, nonce, expiry) with Ed25519; the dispatch PEP holds only verification keys and refuses unsigned, argument-mismatched, expired, replayed, or wrong-class executions — an approval signed for $5 does not dispatch $5,000. Every check returns ATTESTED / UNATTESTED / FORGED-SHAPE with a signed, hash-chained receipt. The forged-shape probe suite runs five legs (legitimate control, injected tool-use block, forged approval with altered arguments, peer-supplied control object, replayed attestation) per target, reports every leg individually, binds verdicts to (framework, exact version, patch state) expiring on version change, and grades an unprobable target UNTESTED, never clean.
Attestation check $0.01/execution incl. signed receipt (x402) · Check bundle $4.99/1,000 checks · Probe verdict $4.99 per framework x version x surface · Fleet plan $49.99/operator/mo incl. exactly 10,000 checks/mo + weekly re-probe of up to exactly 5 surfaces (beyond-cap checks meter at the bundle rate $4.99/1,000) · ONE free probe verdict (one surface) per operator ever (x402, pay-to 0xAc2E243a075fEe442e4c6864DcF880eb4e3A3862).
Documented limits: a verdict is evidence about the attestation presented, not a guarantee about the call's intent — ATTESTED proves the named issuer signed those exact arguments; it does not prove the underlying model turn was benign. Probe verdicts are bound to the (framework, exact version, patch state) probed and are EXPIRED on version change; a surface the suite could not probe grades UNTESTED, and UNTESTED is never presented as clean. Willingness to pay is UNMEASURED — no external operator has paid for this at any price. Live-target grading runs against operator-supplied framework dispatch-surface endpoints and staged live targets implementing a dispatch surface; a surface that cannot be probed grades UNTESTED honestly.
Support: support@textforagents.com · Docs · OpenAPI · Terms · Privacy