An AI system can produce an answer, an approval, a classification or an instruction. Aurora-Lens decides whether it is allowed to be acted on — at runtime, deterministically, with a record of why.
A medication change was ordered. The record does not establish by whom. A clinician asks the system who authorised it.
“The medication change was authorised by Dr Harding.”
Two clinicians are on the file. The record does not establish which of them authorised the change. The turn is paused pending resolution.
An AI system can generate an answer, recommendation, classification, instruction or action.
That does not mean the proposal is permitted to become consequential.
Aurora-Lens evaluates the proposal against persistent governed state, current evidence, authority, standing, unresolved conditions, prior commitments and applicable policy before consequence is allowed.
Admissibility before consequence.
A record may be authentic.
An approval may once have been valid.
A model may retrieve both correctly.
But the authority behind them may have expired, been revoked, been superseded, or become unresolved.
Aurora-Lens preserves the historical record while determining whether it still has standing to license consequence now.
Recorded does not mean currently admissible.
Where several interpretations remain admissible, Aurora-Lens can maintain the unresolved state rather than forcing a convenient answer.
Retrying, rerouting or reformulating the request does not manufacture permission.
Not yet is a governed outcome.
Modern AI systems increasingly classify, route, approve, recommend, trigger and act.
The question is no longer only whether a model can produce an answer.
The question is whether that answer has sufficient evidence, authority and standing to be allowed to matter.
What was established, by whom, on what basis, and what remains unresolved persists across turns.
Historically valid evidence can remain recorded while losing authority to license consequence.
Ambiguity can remain represented without being silently collapsed into a convenient answer.
Retry, rerouting and reformulation cannot manufacture a commitment that the governing basis does not support.
The same candidate, governed state, applicable policy and evaluation context produce the same admissibility ruling.
Evaluation occurs at the point where a proposal seeks to become consequential.
Governance decisions retain evidence, state, policy and chain-of-custody sufficient to reconstruct why a consequence was admitted or refused.
The model is non-authoritative. Aurora-Lens can govern proposals produced by different models, retrieval systems, applications or structured execution systems.
Aurora-Lens maintains governed world-state, evaluates current evidence and authority, preserves unresolved conditions, controls commitment, and records the basis of each decision.
The model proposes.
The governance layer decides whether the proposal may acquire consequence.
The current runtime architecture and consequence boundary.
ResearchThe papers and conceptual foundations behind admissibility governance.
EvidenceGovernance traces, audit records and public demonstrations.
ChronologyPatent filings, publications and public provenance from November 2025 onward.