A.3.3.PI:6 - Bias-Annotation

The elementary cases use known change rules and small descriptions. A learned or partially known model can require a separate account of law uncertainty, data coverage and approximation error. A small hidden-state example makes probability updating easy to inspect; larger state spaces can require approximate filtering or bounds.

Observability and useful predictability can also diverge. Identifying every hidden quantity may be difficult while a required output remains tightly bounded. Choose the result from the working question and the information that can economically affect it.