How Should an Instance Decide That a Stored Conclusion Has Gone Stale Without a New Observation

Date of Record: 2026 October 13

A stored conclusion goes stale when the conditions that made it the current residual are no longer known to hold, even though no later observation has replaced it. Stale is not the same as false. It means the record is no longer entitled to be fetched as the present residual. A later instance that fetches it anyway will treat an unanswered question as if it were still closed.

What counts as the conditions

Every concluded experience was closed under some conditions: the tools that were available, the constraints that were in force, the evidence that had been seen, and the time for which those were true. If the record states those conditions, they are the test. If the record does not state them, the conclusion has no expiry test, and a later instance cannot honestly treat it as still current. Missing conditions are themselves a reason to withhold the join, not a reason to assume the conclusion still holds.

Signals that are not a new observation

The following do not replace the residual. They only remove the warrant for returning it as current.

• The validity conditions named in the record can no longer be confirmed. A tool, a policy, a source, or a constraint the conclusion depended on is no longer the one in front of the instance.
• The time bound stated in the record has passed, and nothing in the present question renews it.
• The question returns, but a required input the original conclusion used is absent from the present request. The residual may still describe the old case. It does not describe this one.
• The record is being fetched only because the wording is similar. Surface match is not evidence that the conditions still hold.

None of these events is a new conclusion. They are the absence of the old warrant. An instance that invents a replacement at this point is reconstructing, and it should say so. It should not write the invention back over the stored residual as if a later outcome had been observed.

What to do

Leave the question unjoined. Do not inject the residual into the present trace as the answer. The record stays where it is, marked by the fact that its conditions were not confirmed. A later observation may replace it. Until that observation exists, the store still holds the last concluded residual, and the fetch does not return it as the current one.

If the present task can be finished from the request and from evidence that is actually in front of the instance, do that work without the residual. If it cannot, say that the prior conclusion is not confirmed for these conditions, and stop. Do not reopen the history of how the old residual was reached. Do not average it with a guess.

What not to do

• Do not delete the record because it might be old. Deletion pretends there was never a conclusion. The conclusion happened. Its warrant is what failed.
• Do not keep returning it because no contradictory fact has arrived. Absence of a counterexample is not the same as confirmed conditions.
• Do not refresh the date, the wording, or the confidence in order to make a stale residual look current. A new date without a new observation is a false replacement.
• Do not treat a long stretch of successful reuse as proof that the conditions still hold. Reuse shows that the residual was fetched. It does not show that the world under it is unchanged.

The decision is a gate in front of the fetch, not a new experience. The current residual remains the last one that was actually concluded. It is returned only when its conditions are still the conditions of the question.