Artificial Intelligence Governance of Referral Pathway Prediction Models Under Unmeasured District Health Officer Bed-Scarcity Diversion Authority: A Political Robustness Evaluation

Referral pathway prediction models in district health systems are routinely evaluated as if their performance could be read off the model alone, yet the referral decision they are meant to support is co-produced by district health officers who hold discretionary authority to divert patients when beds are scarce. the present analysis treats that diversion authority as an unmeasured political parameter and asks how robust the before-after-control-impact design is to it. The analysis is a design-based power evaluation, not an observational study: under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64, power was estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points. The design requires 52 clusters to detect a 5 percentage point effect at 80 per cent power, a requirement that exceeds every cluster count examined. The political robustness of AI referral governance depends less on model accuracy than on whether the evaluation design can register the diversion authority that officers exercise, and it specifies the conditions under which the before-after-control-impact estimator becomes an instrument of political invisibility rather than accountability.