State and Source-of-Truth Control Surface
The State / Source-of-Truth Control Surface defines what information is authoritative for an agent’s action. It separates verified state from memory, stale context, generated assumptions, inferred state, and incomplete retrieval.
What question does it answer?
What source of truth governed the agent’s action?
What does source-of-truth failure look like?
Common failures include acting on an old repository state, relying on conversational memory after context compression, presenting generated citations as retrieved authority, applying a stale policy version, or treating an unverified tool response as authoritative.
The failure is not merely that information was wrong. It is that the workflow could not establish which state governed the action and why that state deserved reliance.
What evidence should an audit inspect?
- versioned or timestamped authoritative sources;
- repository, policy, database, or document identity;
- retrieval results and their provenance;
- distinctions between generated and retrieved content;
- context-compaction and handoff records;
- source freshness and conflict handling; and
- stop behavior when authoritative state was unavailable.
What is the decision consequence of a gap?
Without source identity, an investigator may establish what the agent said but not whether the action was grounded in the state that actually governed the decision. Verification and attribution claims may therefore remain limited.
Related Terms