WHEN AUDIT FINDINGS CONCEAL SKILL GAPS: JOINT VERIFICATION AND TRAINING UNDER A SHARED TIME BUDGET

Audit findings depend on both the occurrence of irregularities and the ability to detect them. This creates an identification problem when the same findings guide investment in auditor skills. We formulate a two-stage decision model in which blinded verification, routine auditing and training compete for fixed novice and expert time. A constructed pair of environments has identical single-review finding rates but different returns to training. We derive the resulting information-limited regret, specify the assumptions under which paired reviews distinguish the environments, and solve the finite verification-budget problem by exact multinomial enumeration. In the nominal benchmark, with 100 future cases per domain, two paired reviews increase expected unique detections from 52.680 to 52.903 while reducing distinct-case coverage from 226 to 224. Verification is not selected at shorter workloads, and the single-pair benefit reverses under modest conditional dependence between reviewers. Twenty thousand Monte Carlo episodes check the exact calculations without serving as additional empirical evidence. The contribution is a decision-focused benchmark linking imperfect detection to costly skill investment, not a new capture–recapture estimator or a field-validated personnel policy. Repeated adaptive allocation and institutional calibration remain outside the implemented model.