All publications
8 publications, newest first.
- Agent Memory Is Not Operational StateWhy durable agent memory, RAG corpora, and knowledge graphs are not the same as formally accepted operational state with sources, authority, and disclosure boundaries.
- Managing Revocation and Conflicting ClaimsHow late evidence, revocations, and conflicting claims change current state — with an illustrative walkthrough and explicit limits on what Tinor claims today.
- Why Agents Need Verifiable StateFrom procurement questionnaires to external AI inquiries: why agents need attributable, current, permission-aware state — and what cryptography alone cannot prove.
- Beyond Ontology: From Understanding Reality to Relying on ItOperational ontologies made business reality computable and actionable. Cross-party reliance asks a further question: is this specific claim admissible, by whose authority, and as of when?
- Proof Is Not TruthWhen a record, credential or signature verifies, what exactly has been verified? Integrity, signer identity, authority, status, scope and applicability are different checks — and none of them is truth.
- An AI Answer Is Not a FactWhat has to happen before a probabilistic interpretation becomes accepted, reusable state: the Fact Admission Boundary, step by step, with a certification claim extracted from a website and a PDF.
- Reality Needs a Type SystemCan real-world claims be compiled into representations with machine-checkable constraints before anyone accepts or acts on them? An engineering note on subjects, scope, time, authority and evidence as types.
- Toward Zero Invalid State Admissions: A Benchmark ProposalA proposed benchmark for stateful AI systems: measure invalid admissions and useful task completion together, so that neither reckless acceptance nor blanket refusal looks like reliability.