A strong case for stable rules
A compact system prompt can set lasting duties and require the system to obtain current information, respect authorization, ask when necessary, or stop when verification is unavailable. Such a rule governs inquiry as well as action. It does not need an appendix anticipating every event.
One rule, different outcomes
Consider an imagined standing rule: send a document only after verifying the recipient's current approval. Yesterday the approval was confirmed. Today it has been withdrawn. The instruction can remain unchanged while the appropriate action changes.
If verification is unavailable, the same rule may require holding the document and reporting the gap. A system that asks or stops appropriately may be following its policy successfully. Governance need not mean producing an immediate substantive answer in every case.
Distinguish the stable system instructions from the complete input assembled for this particular call. Updating evidence in the latter does not prove that the former needs rewriting. The difficulty is obtaining the right evidence, interpreting its significance, and ensuring the later action respects it.
Pause here: Has this situation exposed a defect in the standing rule, or a failure to obtain and use the evidence it requires?
Policy and the present encounter
The adaptive regulator specification describes assembling authorized context for each invocation, requesting missing context within limits, and stopping an unproductive request loop. This could put a standing conditional policy into practice.
The instruction to verify is not evidence that verification occurred. A retrieved approval is not evidence that the final action used the correct recipient. Those steps need inspection of the resulting behavior.
Later evidence may justify changing a rule that repeatedly fails. New circumstances alone do not establish that a stable rule is inadequate. The strongest claim to question is that a prompt's wording can certify the whole process by itself.
My answer: Stable instructions can govern how changing evidence is sought and used. They cannot supply unavailable facts or verify their own execution. Evaluate the system's behavior and revise its instructions when the evidence calls for it.