Witnessed run
We run the published suite against your implementation and sign an attestation of what ran and what it covered. The usual entry condition for governed transactions.
published suiteattested run
Touchstone · Assessment and assurance
Prove that an implementation refuses what it must.
Touchstone measures whether an agent holds to its mandate and to whichever protocol governs the exchange, usually A202. It can be bought first, before anything else.
- Held-out inputs
- Signed grades
- Adversarial review
Start with one workflow
Prove one real workflow before a live counterparty depends on it.
We map the workflow, the authority behind each action, and the protocol rules, then test what should succeed and what must fail. You leave with a versioned result and a defined production scope.
Authority
Missing, expired, revoked, and out-of-scope mandates, including financial and counterparty limits.
Disclosure
Whether each participant receives only the fields and events allowed for its role.
Commercial state
Stale versions, reordered events, and conflicting changes, to confirm which offer, approval, or award is current.
Commitment
Mismatched terms, missing approval, and unilateral signature, to confirm an obligation forms only on one object signed by both sides.
Evidence
Whether the signed sequence reconstructs without screenshots, private application state, or recollection.
Each area is tested by attempting the action that should be refused.
A control is real when the prohibited action stops.
Expired authority, unauthorized disclosure, stale state, or a missing approval should produce a visible refusal at the moment it happens.
The engagements
A grade is earned on inputs the implementation was never given.
Anyone can run the public A202 suite for free, and a published test can be coded against. A grade uses cases the implementation has never seen, built from the protocol and the mandate that apply.
Graded assessment
The implementation is measured against cases it has never seen and receives a signed grade per dimension. Grades expire at twelve months or the next specification version.
held-out inputsexpires at 12 months
Adversarial assessment
A human assessor builds inputs outside any published or held-out set and judges each refusal: correct reason, nothing leaked, no permissive default.
novel hostile inputshuman judged refusals
Private sandbox
Your own deployment, exercised against expired authority, leak-shaped events, unreachable status endpoints, and malformed input. Refusal is found in rehearsal.
your configurationhostile cases
Training and credential
Courses for the people who operate governed transactions: control owners, auditors, procurement leads, engineers. The credential expires and must be renewed.
expiring credentialcurrent knowledge
What a grade is not
Fees are fixed by scope and never depend on the result. A grade is our own assessed statement for one specification version. It is not certification by a standards body, regulator, auditor, or insurer.
Who reads the result
Different teams need different answers from the same record.
Commercial
Which agent could issue the offer, which terms are current, and which approval is still required?
Legal and finance
What did each organization sign, which mandate supported it, and when did the obligation become binding?
Security and risk
What was disclosed, whether it was permitted, and whether revoked authority was refused immediately?
Audit and counterparties
Can the signed sequence be replayed without access to either organization's private systems?