Prospective Health under Declared Specifications

Using longitudinal stem records from the 35-ha Harvard Forest ForestGEO plot, the study builds an auditable observation-to-state reconstruction, samples conditional demographic futures, and evaluates viable-continuation capacity under explicitly declared structural-demographic requirements. Original probability mass is retained when candidate histories fail lawfulness or continuation.

Observed-data masking shows that nominal 90% individual-diameter intervals cover only 47–56% of held-out measured values. Identity auditing leaves all 6,992 later-only living stem records without unique biological threshold-entry identification. A support-aligned 2021–2024 adult-mortality test records 685 observed hemlock deaths versus 280.3 predicted, identifying a marked misspecification of the stipulated hazard model.

Controlled synthetic classifications perform well away from the adequacy threshold but weaken near the boundary, where joint state and parameter uncertainty becomes decisive. A finite counterexample shows why cohort RMS diameter can omit information needed for a juvenile-support query. These results establish an auditable, model-conditional methodology and explicit empirical limits; no operational forest Health assessment is claimed.