State-Dependent Observability in Progressive Systems

In progressive systems, measurements are often computed from components that remain observable even as progression preferentially removes those most affected. Under these conditions, summaries calculated from the observable system may no longer target the same estimands as summaries defined over the complete system. Using a generative simulation framework, we show that state-dependent observability progressively depletes high-state components from the observable system, inducing systematic divergence between observed-system and complete-system estimands. This divergence is estimand-specific: summaries drawing information from the upper tail of the state distribution exhibit substantially greater attenuation than whole-distribution summaries. The pattern is structurally general, arising across alternative observability functions, heterogeneity structures, and coupling strengths, indicating a general measurement problem rather than a model-specific artefact.