Provable decision infrastructure for human, automated, and AI-assisted operations.
Modern operational systems are becoming increasingly automated, interconnected, and AI-assisted. The accountability burden does not disappear when the systems change.
Most institutions still reconstruct consequential decisions retrospectively from fragmented operational evidence. MeshQu preserves independently verifiable accountability records at the moment decisions occur.
Audit becomes retrieval, not reconstruction.
Tuesday, 14:22 UTC. A fraud rule fires, an override is set, the transfer goes through. The system moves on.
“Why did this decision pass? What did the system actually see at the time?”
None of these were created as proof. All of them depend on memory, access, and time.
Best reconstruction. Plausible narrative. Not authoritative, not verifiable, not portable. The decision is defended, but never proven.
One signed object, issued in ~50 ms, replayable years later — independent of the systems that produced it.
We hold the institution's policies — the thresholds, the authority levels, the conditions, the exceptions — in a form that doesn't drift. When a decision is made, we know exactly which version of the rules was in force at that moment.
We run the decision against those rules at the moment it happens. Who made it, what they saw, what the rules said, what the outcome was — all of it captured in one place, in one shape, whether the decision was made by a person, a workflow, or an AI.
We emit a signed receipt — a portable, replayable record that holds the decision in the form it was made. Years later, when someone asks how it happened, the answer is retrieved, not reassembled. The proof outlives the systems that produced it.
Three things, one shape — every consequential decision the institution makes.
A Decision Receipt preserves the record of the decision itself.
The systems can change. The proof remains.
A receipt does not prove that the decision was correct, that its inputs were true, that the outcome was fair, or that it was lawful everywhere. It does not prove that a named human personally acted, or that a model followed the rationale it stated.
The same shape regardless of which actor produced it. A receipt records who or what decided, under which policy, on what evidence — and the object is identical whether that was a person, an AI system, or a platform you already operate.
A valid signature shows that a trusted key signed this exact receipt and that its signed contents have not changed. It does not show that the decision itself was correct or that its inputs were true.
Mechanically, the integrity hash binds the receipt’s contents and the signature binds that hash, the timestamp and the key identity. A verifier checks both, and recomputing the hash is the step that catches an edit.
A velocity-anomaly alert escalates from screening to human review to override. Each step seals into a replayable accountability chain.
A model recommendation is bound to the policy snapshot it was evaluated against. The human reviewer who ratified it is named in the record.
A procurement workflow moves from spec to quote to review to approval, then closes with a chain seal — the sequence is verifiable end-to-end.
Agentic AI governance ends up being about a boundary rather than a model: the point where an agent proposes an action and something decides whether it runs. A receipt can carry what happened at that boundary. It cannot tell you which agent was standing there — and that limit is worth stating plainly.
Over the past few months, three independent constituencies — a European policy process, UK public opinion, and a Singapore legal working group — have reached for the same requirement in three different vocabularies. They agree on the need. None has yet said what form the answer should take.
MeshQu uses anonymous analytics (Vercel Web Analytics) to understand which pages are useful. No cookies, no personal data, no advertising trackers. See our privacy notice for the full picture.