Artificial Intelligence replication of District Hospital Oxygen Concentrator Fleet Rotation Under Unmeasured Sieve Bed Saturation Cycling: A Healthcare Scheduling Robustness Evaluation

District hospital oxygen concentrator fleets are rotated on schedules that assume sieve bed saturation can be inferred from run-hours, yet saturation cycling is rarely measured directly and rotation policy is therefore set against an unobserved state. the present analysis treats fleet rotation as a scheduling problem under partial observability and evaluates the robustness of a linear rotation estimator when the saturation state is unmeasured. A replication-based evaluation was conducted: 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 behaviour of the ordinary least squares estimator was characterised across replications. The estimator recovered the exposure coefficient with estimated bias of -0.00032 and RMSE of 0.01617, while the intercept showed estimated bias of 0.01087 and RMSE of 0.87329; empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. The results indicate that a correctly specified linear rotation model is unbiased and well calibrated for the exposure effect even when the underlying saturation state is not observed, but that intercept-level quantities carry substantially greater sampling uncertainty. The contribution is a bounded robustness result: rotation schedules keyed to the exposure coefficient can be defended under unmeasured saturation cycling, whereas schedules keyed to absolute saturation thresholds cannot, because the intercept is the less stable target. The paper concludes that fleet rotation policy should be expressed in relative rather than absolute terms when sieve bed state is unmeasured.