Artificial Intelligence Benchmarking of Ensemble Rainfall-Driven Cholera Forecast Models Under Unmeasured Latrine Coverage Reporting Lag: A Healthcare Decision-Value Evaluation

Cholera forecast models that combine rainfall drivers with ensemble machine learning are increasingly proposed as decision aids for outbreak preparedness, yet their evaluation typically assumes that sanitation covariates are observed without delay. The framework formalises reporting lag as a latent state that decouples the covariate process from the decision time, defines a lag-robust ensemble aggregator, and derives a decision-value bound that separates forecast discrimination from the economic cost of acting on stale sanitation information. Using NASA POWER monthly reanalysis air temperature at 2 m for Kampala (0.35, 32.58) over 2000-2023, comprising 312 monthly observations with a mean of 22.443 C and a linear trend of -0.1564 C per decade, the paper demonstrates how the benchmarking protocol is operationalised on a real covariate stream. The contribution is a reproducible evaluation design that reports decision value rather than accuracy alone, and that makes the reporting-lag assumption an explicit object of benchmarking rather than a hidden modelling convenience.