Researchers have developed an auditable framework for trustworthy large language model analytics in enterprise finance. The Knowledge-Driven Analytics Framework (KDAF) builds ontology-driven knowledge systems through six iterative stages and retrieves evidence via Context-Aware Relevance Propagation (CARP). This allows every retrieved fact to carry its relationship type, confidence, and source lineage. An evaluation on FinanceBench showed that KDAF outperforms other methods in terms of auditability, with the highest citation traceability F1 score. However, accuracy alone does not justify structured retrieval in this context. --- Why it matters: This matters because enterprise finance applications require answers to be both accurate and auditable, as they must be traceable to authoritative sources. The KDAF framework addresses this need by providing a way to evaluate and trust large language model analytics in regulated workflows. Source: https://arxiv.org/abs/2608.20661