Trust
Trust by architecture, not by assurance.
We would rather show you a structure that makes a failure mode impossible than write a paragraph promising it will not happen.
- Least authority
- Exact evidence
- Separated duties
Commitments
Six properties the system is built to hold
Each one is enforced by where authority sits, not by policy or intent.
- Least authority
- Every component gets the narrowest authority that lets it do its job. The verifier cannot deploy. The signer cannot propose work.
- Explicit consequence envelopes
- Scope, budget, blast radius and expiry are declared before admission. Out-of-envelope work is refused, not renegotiated at runtime.
- Exact-commit evidence
- Evidence binds to a specific commit or prepared state, never to a summary written afterwards by the component that did the work.
- Separation of duties
- Proposer, authority, executor, signer and verifier stay distinct roles. Collapsing them collapses the proof.
- Deterministic recovery
- Leases, idempotency keys and durable transitions mean an interrupted run resumes to a known state instead of repeating side effects.
- No self-verification
- A run does not close because the agent reported success. It closes when an independent attestation says state matches claim.
Boundary
The one line the system will not cross
Verification sits outside the execution path. Evidence crosses the boundary; authority never does.
Execution side
may act
- Inputobjective
Human objective + authority
A prose outcome and an explicit authority envelope enter the system together.
- Execution / control planedonestate
DoneState
Admits the run, holds the lease, records durable transitions, halts at AWAITING_VERIFICATION.
- Authorised transactionsagentproof
AgentProof
Consequential effects are bound to prepared state and emit a signed receipt.
Verification side
observes only
- Surfaceevidence
Observable evidence
Repository at an exact commit, CI output, configuration, routes, runtime and deployment artefacts.
- Independent verifieropstruth
OpsTruth
Reads only. Classifies each check as Verified, Risky or Unproven.
- Outputattestation
Attestation closes state
The run closes when an independent attestation says observed state matches the claim.
Self-verification is not proof
Principles
The positions behind the commitments
Three long-form technical positions. They are arguments, not slogans, and they are meant to be disagreed with in detail.
Principle 01
Independent verification
A claim of completion is evidence about the claimant, not about the system. Verification only means something when the verifier is structurally incapable of having caused the outcome it judges.
Principle 02
Least authority
Every component receives the narrowest authority that lets it do its job, declared before it starts and enforced at admission rather than argued about at runtime.
Principle 03
Evidence over claims
Prose summaries decay, get rewritten and cannot be re-checked. Evidence is bound to an exact commit or prepared state and can be inspected long after the session that produced it has ended.
What we do not claim
Maturity is not a marketing decision.
- We do not list customers, logos, adoption counts or partnerships. None are claimed because none are being asserted here.
- We do not claim certifications, audits or compliance attestations. If that changes, the evidence will be published alongside the claim.
- We do not publish uptime figures or live service metrics. See status for what is actually known.
- We do not list releases that have not been cut. The changelog stays empty until there is something real in it.
- Definition — Proof & State
- Proof & State is AI work accountability infrastructure: DoneState executes under declared authority, AgentProof authorises consequential actions and signs receipts, and OpsTruth verifies the result read-only from outside the execution path.
Give agents authority. Keep the proof.
Every commitment on this page should be checkable against the source.