Model validity is a resource, too.
The framework introduces epistemic resources: models, estimators, identifiers and learned policies whose validity must be monitored just like the physical state.
When not the sensor but the knowledge degradesExpand section
When not the sensor but the knowledge degrades
The system may lose not only a physical element but also the validity of its own knowledge: an outdated model, an incorrect estimator or a corrupted learned policy.
This is dangerous because hardware may still look healthy while decisions already rely on a false basis.
How an epistemic resource is evaluatedExpand section
How an epistemic resource is evaluated
An epistemic resource receives a validity coefficient that decreases with prediction error and the time since the last validation.
This validity is subject to the same supervisor decisions: the model may be deactivated, replaced, revalidated or compensated by another configuration.
Restoring validity is also an actionExpand section
Restoring validity is also an action
Once epistemic resources are introduced, reconfiguration includes not only switching hardware or software modules, but also actions that restore knowledge: recalibration, re-identification, activating a reserve model, or falling back to a simpler but still valid control law.