Artificial Intelligence Benchmarking of Oral Rehydration Solution Stock-Out Forecasting Under Unmeasured District Store Ledger Backdating: A Healthcare Supply-Robustness Evaluation

Stock-out forecasting for oral rehydration solution (ORS) at district stores is routinely evaluated with before-after-control-impact (BACIP) designs, yet the ledger records that supply these evaluations are frequently backdated, and the resulting bias is rarely quantified. the present analysis treats the problem as a design-characterisation question rather than an empirical one: given a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.640), what are the operating characteristics of the BACIP estimator when the outcome is a forecast-error metric rather than a stock count? A replication-based evaluation was conducted across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. Power rose with cluster count and effect size, and 56 clusters were required to detect a 5-percentage-point effect at 80% power. The minimum detectable effect at 80% power fell monotonically as clusters increased. The intraclass correlation of 0.640 dominated the variance budget, and the BACIP interaction estimator remained approximately unbiased under the declared structure but widened sharply at 8 and 16 clusters. The contribution is a design-characterisation framework that separates the ledger-backdating problem from the estimator problem, and that specifies the cluster counts at which ORS stock-out forecasting evaluations can and cannot support robust inference.