The solution

Trust, but verify. ProofSmith is the verify.

ProofSmith is the trust layer for digital information. It establishes provenance, decides what may access or transform an artifact — permit or deny, on authority — and emits independently verifiable proof of that decision, whether the actor is a human, an AI, or a machine.

The architecture

Three things we claim of it, and each can be tested.

We took the founders’ answer and built it for the Age of the Machine. Nothing on this site describes how the architecture is built. What it does, at the altitude of a claim you can check, is this.

The decision is adjudicated before the act. The proposer proposes. The Arbiter decides. Before an action runs, it checks the request against the authority for it, applies the authority you delegated rather than a judgment of its own, and signs a permit or a deny. The Arbiter does not make a model more accurate, and it does not judge whether the authority behind a decision was sound. It rules on one thing: whether the act was inside the authority you wrote.

Where the adjudicator runs. The adjudicator has to run somewhere the actor cannot reach. That is a property of the platform, and one the customer can check for itself.

The proof is checkable afterward by anyone you authorize. Anyone you authorize can check that record later, using methods we publish to them, and reach the same answer without trusting us, the vendor, or the system that acted. Of the seven events on AI is Running Wild, the three in which an act ran would each have left a record that did not depend on the actor.

Where every other system stands against two tests, and the count, is on Category Definition. The requirement behind each claim is on The Problem.

The decision path

We sit between the request and the action.

Not a new decider. The authority you wrote, enforced in-line at machine speed. You author the authority; the Arbiter enforces it and proves every decision faithful to it.

Any proposer

Proposes an action — not yet executed.

Human
Machine
AI
SI

The Arbiter decides

Evaluates policy, decides PERMIT or DENY, signs.

Authority
User’s intent
User’s rules
Every party, every step

Communication capsule

The signed permit travels with the request.

Requested action

Runs on a permit. Does not run on a deny.

Evidence artifact

Action, policy, authority, outcome. Signed, hash-chained, tamper-evident.

Any authorized verifier

Reproduces the result from the signed record alone, no vendor trust.

Each artifact references its authority — decisions chain into a verifiable lineage up the chain of authority.

The descriptor

Independent & Provable GOVERNED DECISION LINEAGE

Five words. Each one is a claim, and each one can be tested.

Independent

Separate from the system that acts and from the party that built it, and checkable by a party that needs nothing from either. The independence is of the judgment and the proof; the authority stays yours.

Provable

Checkable by a party outside the system, using methods we publish to the parties you authorize, without the cooperation of the system that acted. Self-attestation is not proof.

Governed

The action was permitted, not merely observed: decided against authority written in advance, before it ran, and denied if it fell outside that authority.

Decision

Not logs. Not intent. The exact action the agent tried to take, the policy that applied, and the outcome.

Lineage

The chain from authority through decision to action to evidence, unbroken, so that it can be examined in audit and in litigation.

What the architecture produces is a record of every permit and every deny, signed at the moment of decision, that anyone you authorize can verify later without asking us or the system that acted.

The demonstration

Tell us which of the six is costing you.

Start a conversation See the six problems Download the two-page summary