In deployed systems a control policy can hide consequential dynamics changes. For instance, an actuator may lose effectiveness but, because the policy rarely excites it, the current task remains unaffected while a future, as‑yet‑unspecified task could fail catastrophically. To address this we introduce the decision problem of task readiness under dormant dynamics drift, which unifies active diagnosis and post‑change recovery within a limited, task‑agnostic interaction budget. An agent must decide whether local dynamics have changed, locate the change, and use a small number of informative interactions to characterize it before the downstream task identity is revealed. Afterwards, for each candidate task the agent either supplies a recovered policy together with a calibrated lower bound on achievable return, or abstains in favor of a safe fallback. We propose Evidence‑Gated Matched‑Pulse Transport (EG‑MPT), an intervention‑based Bayesian procedure that jointly performs fault localization and actuator‑effectiveness estimation via a shared matched‑response representation, preserving diagnostic reliability while converting localized evidence into recovery‑relevant uncertainty. This uncertainty is propagated to task‑conditioned policy selection and readiness certification, enabling deployment decisions that explicitly trade off expected performance, confidence, and fallback usage. We evaluate the framework on a diverse suite of dormant‑actuator benchmarks across multiple simulators, using a protocol that separates diagnosis from capability recovery and scoring deployments by readiness coverage, selective risk, interaction cost, and return, while identifying fault regimes where transported evidence is decisive.
Review