Universal Dot System / Technical
Explicit states.
Separate authority.
UDS applies established reference-monitor principles to AI epistemic state and action authority. It does not claim to control what an AI thinks.
01 / Claims
A claim has a state and a history.
A claim’s status describes its support under the system’s rules. Progression is governed; it is not an automatic ladder to truth.
- UNKNOWN
- The claim has not yet been established.
- INFERRED
- The claim is an inference from available information.
- ASSERTED
- The claim has been explicitly put forward; assertion alone is not verification.
- VERIFIED
- A claim-state label reserved for acceptance under the designated evidence and verification-authority rules.
- STALE
- Earlier support requires re-evaluation before further reliance.
VERIFIED is a UDS state. It is not a guarantee of universal truth, model correctness or independent security validation. Claim status also does not, by itself, authorize an action.
02 / Actions
Capability does not confer permission.
The agent proposes an action. The governed path evaluates evidence and authority under explicit rules. An allowed request may proceed through the mediated execution path; blocked or quarantined requests do not.
A receipt records the decision and observed outcome. A receipt can make a transition inspectable; it cannot demonstrate the absence of every possible bypass.
- AgentProposes a claim or action
- GovernanceEvaluates evidence and authority
- Execution boundaryMediates the allowed action
- ReceiptRecords decision and outcome
03 / The hard boundary
Only the mediated path is governed.
Actions outside its mediated path are not controlled by UDS.
A general-purpose agent may have shell access, credentials, mutable execution surfaces and network egress. Complete mediation requires that the relevant consequential actions cannot bypass the control boundary.
The bounded prototype test record does not establish that condition in a real deployment. Provider-credential custody, non-bypassable egress, transition enumeration and independent adversarial validation remain material open work.
UDS applies existing security concepts to claim promotion, provenance, verification authority, acceptance authority and action authority. This is an application claim, not a claim to have invented reference monitors or complete mediation.
The proposed first application is inspection work-order closure, including a tri-drone inspection workflow. The browser example illustrates missing inspection evidence and absent or revoked closure approval. No field integration or operational pilot is established by this site.