ABSTRACT This study examines how students develop professional judgment by analyzing the “Income Taxes” footnotes of Apple Inc. and Amazon.com Inc. The design guides a four-step workflow: (1) compute effective tax rates; (2) identify qualitative drivers: unrecognized tax benefits, deferred tax assets/liabilities/, valuation allowances, and cross-border/legal exposure; (3) apply a deterministic large-language-model prompt using context engineering; and (4) reconcile human and model conclusions to deliver a concise evidence-based comparative conclusion. Structured worksheets and reflections promote reproducibility and professional skepticism. The paper further provides validated reference analyses, additional guidance, and discusses retrieval-augmented generation (RAG) as an optional pathway for source validation and auditability. By linking computational tools with evidence-based reasoning, the study provides a rigorous, ready-to-use framework that accelerates the first-pass reading of dense disclosures while maintaining a central role for human judgment.
Lin et al. (2026) studied this question.