Digital twin–based structural health monitoring increasingly relies on sequential estimation of structural states, unknown inputs, and time-varying parameters from sparse and noisy measurements. While reduced-order models combined with Bayesian filtering enable real-time estimation, stable convergence of estimated parameters does not necessarily imply physically meaningful or uniquely identifiable structural changes. This work presents an uncertainty-aware interpretation framework for sequential stiffness estimation in reduced-order digital twins, aimed at distinguishing detectable parameter evolution from uniquely identifiable structural change. A parametric reduced-order structural model with clustered stiffness parameters is embedded within an augmented-state Extended Kalman Filter to jointly estimate structural states, unknown excitation, and stiffness evolution using limited strain and acceleration measurements. Stiffness parameters are modeled as stochastic states without prescribing explicit degradation laws, allowing autonomous propagation during measurement dropout while explicitly representing loss of confidence through covariance growth. An interpretation criterion based on the joint evolution of parameter estimates and their associated uncertainty is introduced to distinguish identifiable parameter changes from uncertainty-dominated variations, building upon recent joint state–input–parameter estimation frameworks. Numerical investigations on representative structural systems demonstrate that parameter uncertainty may contract even when spatial localization remains non-unique due to reduced-order sensitivity coupling, sparse sensing, and limited excitation richness. Measurement dropout scenarios further show that uncertainty growth provides a consistent indicator of diminishing identifiability in the absence of informative data. The results highlight that convergence of sequential estimators alone is insufficient for reliable structural interpretation and that uncertainty-aware analysis is essential for conservative and physically meaningful digital twin–based decision-making.