About
An agent said it was done. That is not the same as it being done.
Proof & State exists because autonomy scaled faster than accountability. The tooling to let an agent act is mature; the tooling to prove what it actually changed is not.
Every coding agent already produces a completion signal. The problem is what that signal is made of: the same model, the same context and the same assumptions that produced the work also produced the report about it. When the work is wrong, the report is usually wrong in the same direction.
The response has mostly been to make agents more careful. That helps, and it does not solve the structural issue, because no component can supply independent evidence about itself regardless of how careful it is.
Proof & State takes the other route. Split the system so that the part that acts and the part that judges are different components with different authority, then make the artefacts durable enough to inspect after the session ends.
That is the entire thesis. DoneState executes under declared authority and refuses to close its own run. AgentProof binds consequential actions to exact prepared state and signs a receipt. OpsTruth reads the result from outside and classifies it Verified, Risky or Unproven.
Definition
What Proof & State is
- 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.
How we work
Source first, claims second.
The repositories come before the marketing. Where this site describes behaviour, the corresponding repository should let you check it.
We would rather publish an empty changelog and an honest status page than manufacture momentum. Accountability infrastructure that overstates itself has already failed its own test.
Give agents authority. Keep the proof.
Autonomy is fine. Unverifiable autonomy is not.