Enterprise Document Intelligence Vol.1 #14D explores why standard retrieval-augmented generation (RAG) fails on case files like insurance claims. A case file is a bundle of unlike documents—contracts, invoices, photos, reports—about one entity. The unit of work is the bundle, not individual files.

Two critical questions illustrate the gap: whether a second repair invoice exists in the folder, and whether the date of loss matches across the claim form and adjuster's report. Neither is a search question; the first concerns a missing document, the second requires comparing values across files. Models read the middle of long contexts less reliably, and in an eleven-PDF bundle, both key values sit in the middle.

The article introduces a completeness index listing expected pieces per case type, enabling detection of missing documents as first-class answers. It also covers cross-document checks needing typed values, and the answer shape: the state of the case, not a passage. This is Part IV of the series, where input becomes a folder behaving as one object.

A companion notebook at doc-intel/notebooks-vol1 runs completeness checks on custom folders, demonstrated on a fictional fire claim at a joinery workshop.