Certified bipartite alignment and cross-database transfer in electronic health record integration

Abstract

Background: three governorate profiling exercises documented displacement in the Kurdistan Region of Iraq, and their household microdata have never been reanalysed. Methods: an instrumented pipeline audits provenance, aligns the two admissible instruments by maximum-weight bipartite matching with a linear-relaxation optimality certificate, and tests cross-governorate transfer of stated non-return. Results: identical labels recover 6 of 84 correspondences; the certified alignment recovers 60 with no false pair and all 84 at a precision of 0.966. Non-return is predictable within a governorate and no better than chance across them. Conclusions: comparability must be computed, not assumed.