Indigenous communities in the United States continue to experience persistent disparities in health and public-safety outcomes shaped by long-term structural and historical conditions. This study presents a source-documented computational pipeline that rebuilds processed analytic tables from raw inputs, constructs a state-level \emph{Structural Exposure Composite} from observable structural indicators, and evaluates its association with selected contemporary disparity measures in an ecological design. The composite is constructed from normalized weighted indicators covering historical-policy presence, environmental burden, and boarding-school context. Mortality data are pooled across years (2019--2022) and conditions (diabetes E10--E14, liver disease K70--K76, suicide) from CDC WONDER, expanding state-level coverage from 13 single-condition states (the original specification) to 19 states with at least one (year, condition) cell after suppression. The complete-case analytic sample comprises 13 states with simultaneous coverage across all input domains. Permutation-based p-values (10{,}000 iterations) are the primary inferential tool; bootstrap confidence intervals (4{,}000 iterations) are reported as descriptive only; Holm-Bonferroni and Benjamini-Hochberg corrections are applied within the primary outcome family; and a Dirichlet weight-perturbation analysis (2{,}000 draws) characterizes robustness to weight choice. \textbf{Headline result.} In the complete-case sample, no primary association between the Structural Exposure Composite and a contemporary disparity outcome attains statistical significance after multiple-testing adjustment. The configured-weight composite is essentially uncorrelated with mean mortality disparity (, permutation , ) and weakly positively correlated with the absolute AI/AN missing-persons count (, raw ), but the latter attenuates to () after adjustment for AI/AN population share. The equal-weight co-primary specification yields the same qualitative pattern. A formal selection-bias audit confirms that the complete-case sample over-represents states with higher AI/AN population (Mann-Whitney permutation for AI/AN population; for AI/AN population share). The principal contribution of the study is methodological: a transparent and auditable workflow that makes raw-data reconstruction, multi-condition pooling, normalization, sensitivity analysis, multiple-testing correction, construct-validity diagnostics, and selection-bias auditing explicit.