Artificial Intelligence replication of District Hospital Snakebite Antivenom Stock Allocation Under Unmeasured Cold-Chain Excursion Logging Gaps: A Healthcare Robustness Evaluation

District hospitals that hold snakebite antivenom must decide how much stock to keep and how to allocate it across presentation severities, yet the cold-chain excursion logs that would justify those decisions are systematically incomplete. the present analysis treats the missing logging as a structural feature of the allocation problem rather than a data-cleaning nuisance, and asks how an allocation rule calibrated on observed antivenom demand behaves when the cold-chain state of part of the stock is unrecorded. A replication framework is specified in which antivenom units carry a latent potency state, excursion events are logged with severity-dependent probability, and an artificial-intelligence allocation policy is trained on the logged subset and deployed on the full stock. The design is evaluated through a replication-based replication study: 600 data sets of N = 500 were drawn from a declared data-generating process (intercept 12.0, exposure 0.42, group -3.1, severity 1.35, error SD 4.2), and the ordinary least squares estimator was characterised under that process. Across the 600 replications the intercept bias was -0.03426 with RMSE 0.89784, the exposure bias was 0.00059 with RMSE 0.01631, and nominal 95% coverage was 1.0 for both parameters. Robustness of an antivenom allocation policy is bounded by the identifiability of the cold-chain state, not by the sophistication of the policy, and it derives the conditions under which unlogged excursions become an irreducible allocation hazard.