Artificial Intelligence replication of District Hospital Insulin Cold-Chain Allocation Under Unmeasured Vial Agitation Exposure: A Healthcare Robustness Evaluation

District hospital insulin cold chains are routinely perturbed by vial agitation that is never logged, so allocation rules are evaluated against a covariate that is measured with error. the present analysis treats that measurement gap as a formal problem in estimator robustness rather than as a data-cleaning nuisance. A declared data-generating process with intercept 12.0, exposure coefficient 0.42, group effect -3.1, severity coefficient 1.35 and error standard deviation 4.2 was replicated 600 times at N = 500 observations per replication (seed 665328261), and the ordinary least squares estimator was characterised across those replications. The intercept was recovered with estimated bias 0.01689 and root mean squared error 0.91435; the exposure coefficient was recovered with estimated bias -3e-05 and root mean squared error 0.01626; nominal 95 per cent confidence interval coverage was 1.0 for both parameters. The sampling distribution of the exposure coefficient was approximately normal, the Q-Q plot was close to the identity line, and the fitted values calibrated against observed values along the 45-degree diagonal. The contribution is a bounded robustness statement: under the declared process, unmeasured agitation exposure of the magnitude encoded in the error term does not by itself destabilise the allocation coefficient, and the binding constraint on district-level insulin allocation is therefore the absence of agitation logging rather than estimator fragility. The paper specifies what a logging regime would have to record for that conclusion to be tested against field data.