E.23.CAE:2 - Problem

Apparent capability loss compresses different failures into one sentence: “they knew it yesterday,” “the organization forgot the routine,” “the model catastrophically forgot,” or “the robot cannot do it anymore.” Each sentence can trigger expensive or harmful Work before anyone checks whether the old response still appears under a qualified condition.

The opposite error is to explain every recovery with one theory. Human sensorimotor memories can be expressed according to contextual inference; conceptual knowledge can remain inert until relational retrieval; organizational performance can depend on roles, rules, records, artefacts, and authority; an AI function can remain represented while its activation is biased; a robot can retain a policy while sensing, state estimation, actuation, or configuration prevents enactment. These structures are not one memory system.

The reusable problem-solving move is narrower: bind one capability claim, control further change long enough to compare conditions, recover a reference response where possible, separate observable positions, retain rivals, and return an evidence-bounded disposition. The method must admit both genuine capability change and an unresolved result.