High-risk artificial intelligence systems are usually validated under a stable-data assumption, even though the distributions that shape predictions and explanations change after deployment. This study examines predictive degradation, explanation degradation, uncertainty, subgroup disparity, and adaptive governance across four temporally ordered decision settings: healthcare utilization, macroeconomic financial risk, energy-climate planning, and public-sector vulnerability prioritization. The empirical design combines real public data with natural temporal shifts in three domains and a documented controlled shift applied to real healthcare covariates. Logistic regression, random forest, and gradient boosting are evaluated in pre-shift, shift, and post-shift windows. Feature-ablation explanations are assessed through fidelity, temporal rank stability, directional consistency, actionability, subgroup disparity, and uncertainty. These components form a six-part Explanation Degradation Index. Static, recalibrated, periodically retrained, rolling, expanding, drift-triggered, and joint model-explanation strategies are compared under blocked temporal evaluation. Results show that predictive and explanation degradation are related but not interchangeable. Mean EDI increased by 0.032 during shift, while cross-domain changes in precision-recall performance were heterogeneous and statistically inconclusive. Recalibration frequently improved Brier score but left EDI unchanged. Joint adaptation with uncertainty-based deferral improved mean precision-recall area by 0.020 across eight domain-period tasks, although pairwise gains were not statistically decisive and finance exhibited a negative shift-period result. The Adaptive Explainability Under Shift Framework translates these findings into monitoring, escalation, adaptation, approval, rollback, and organizational-learning controls. The study contributes a measurable account of explanation reliability under change and a governance design that separates predictive confidence from explanation trustworthiness.