SYSE.46:4.4 - Test persistence and changed conditions when relied on

When the claim includes lasting improvement or response to change, compare initial, post-change, delayed and shifted use. Select the interval and shift from that reliance: normal restart, memory expiry, source revision, tool withdrawal or altered workload. There is no universal waiting period.

At each observation retain the task/result, configuration identity, support actually available, evaluator basis and intervening changes. Test delayed persistence under stable comparison conditions separately from changed support. An unidentifiable model-provider update may leave current system performance observable while making a component-specific persistence claim unresolved.

Keep warranted restraint, unnecessary and missed necessary access, actual obtained-and-used result, verification, sufficient-result stopping and total burden visible. A changed authoritative fact should be sought and used even after repetitive retrieval becomes cheaper. A previously useful operator should be rejected when its applicability is defeated. Removing a necessary aid changes the tested performing arrangement.