We built this before we knew
what we had found.
In late 2025, we were trying to solve a specific problem: how to make an automated decision about a person legitimate at the moment it happens. Every other approach intervenes after the fact. The USS intervenes before. Because of that, every decision it permits is also reconstructible, auditable, and defensible. Not as a feature bolted on. As a structural consequence of getting the timing right.
A decision was made about you. You do not know who made it, what they were authorised to decide, or whether anyone was accountable for the outcome. The system acted as if it had authority when no human ever declared it.
That is Ghost Authority. It remains structurally identical whether it happens in a hospital, a welfare office, a bank, a hiring process, or a care home. The domain changes. The structure does not.
The USS exists to make Ghost Authority structurally impossible. Not as a policy or a guideline. As a gate that either opens or stays closed.
Select your position and a domain below. The gate runs two versions of the same decision: one with declared authority and one without. Watch what the mathematics does to a real decision.
L_valid false
state GAMMA
gate_hash null
error 403-GA
L_valid true
state ALPHA
gate_hash [SHA-256]
replay FIRST_SEEN
What the gate just decided
What declared authority changes
Why this matters to your institution
The USS sits at the boundary.
The AI runs freely inside a case. Nothing leaves the case to act on a person without passing through the gate. The USS does not read the model and does not need to. It sits where the system’s output meets the person it acts on. Any action that alters a person’s situation reaches the gate, whatever its form: a decision, a flag, a reordered list, a draft for a human to sign.
Governance is part of the design and cannot be added afterwards. Before the system runs, each organisation defines its decision classes and names the human who holds authority for each. That definition is theirs: a hospital groups its clinical actions and names who holds each group; a lender does the same for credit. Every class must be bounded. A declaration that covers every decision declares nothing.
At runtime the gate checks every action against that design. The action proceeds only when a named human’s declaration is intact, signed by that human and made for a class they hold, with a contestation pathway, a response obligation and a bounded timeline declared. Otherwise it does not proceed. Every evaluation is sealed into an audit record that must reach the person it concerns.
The gate checks configuration, not content. It confirms that authority exists, not that each decision was correct, so it scales with decision classes, not decisions.
Where authority cannot be confirmed, the gate stays closed and the action waits. That is the system working, not a bottleneck. Reopening is not automatic: the gap must be repaired and authority revalidated for that class.
USS™ v2.6 is a deterministic, rule-based governance engine. Three detection layers. Four structural violation rules. Eleven automated classification categories derived from the ADCI closure across two domains. Six structural invariants. Two hundred and thirty-one tests passing, none failing, across fourteen test files. Every invariant enforced directly in code.
Authority is declared, never inferred. It is sealed at the moment of decision or pre-registered before it, and signed either way. A hash proves the declaration is intact. A signature proves who wrote it. The gate reports those separately, because an untampered record is not the same as one someone stands behind. Each authority holder is enrolled at design time for the decision classes they hold, and a signature counts only for one of those classes. Classes match exactly: no wildcards, no patterns. The governance layer and the value layer stay quarantined from each other, with no configuration under which that relaxes. Some of this is enforced by tests that read the code rather than run it: if a decision engine were ever reintroduced into the routing layer, the build would fail, naming the violation.
Existing governance systems describe what should happen. The USS™ enforces what structurally can happen. It is not a policy layer bolted onto an existing system. It is the layer that determines whether the system has permission to act at all.
The proof runs.
We welcome inquiries into the USS™ from institutions, regulators, and implementation partners.
[email protected]