Artificial Intelligence Governance of Health Facility Catchment Boundary Models Under Unmeasured Ward Delimitation Committee Gerrymandering: A Political Robustness Evaluation

Catchment boundary models used to allocate health facility resources in low- and middle-income countries are increasingly produced by artificial intelligence systems, yet the political conditions under which those models operate remain untheorised. The framework distinguishes three mechanisms through which unmeasured delimitation discretion degrades catchment model performance: unit instability, population denominator distortion, and facility assignment drift. Using the DHS Program indicator 'Nonusers who visited a health facility and discussed FP' across 131 country-survey observations from 36 African countries between 1994 and 2025 (mean 12.674, SD 6.52), the paper demonstrates that the measured variation in facility-based family planning discussion is consistent with substantial unmeasured spatial-political interference in the catchment inputs that AI models would consume. The contribution is a governance framework that specifies when AI catchment models should be certified as politically robust, when they require delimitation-sensitive recalibration, and when they should be withheld from resource allocation decisions pending institutional reform of the delimitation process itself.