Count document families, not just PDFs
Give each manual a family ID and each revision its own ID. Record equipment model, language, issue date, status, and page count. A bundle containing the same manual in several exports is different from several distinct manuals. Keep superseded documents identifiable instead of silently mixing them into the current instructions.
Inventory text, illustrations, and inserts separately
A manual may combine internal writing, vendor diagrams, translated sections, and third-party specifications. Link those components to their source records and unresolved permissions. If the source of an illustration is unknown, identify the affected page range and leave it out of a proposed sample until an appropriate reviewer decides how to handle it.
Test extraction where structure matters
Select pages with tables, multi-column text, diagrams, warnings, and scanned print. Compare extracted content against the page image. Record missing cells, scrambled reading order, lost units, and broken cross-references. A count of successfully opened PDFs says little about whether the extracted instructions still make sense. Keep OCR output linked to its original page and revision.
Make applicability part of retrieval evaluation
Use example questions that require a model or revision distinction. Ask the evaluator to show which source revision supports an answer and how it handles outdated instructions. This is an illustrative test design, not validation that the material is safe for operational use. Domain specialists should define any safety-related evaluation and deployment boundaries.
Deliver a revision map with the files
Provide document_id, revision_id, supersedes_id when known, product applicability, language, and page-level source references. Record unresolved links and missing revisions. If the proposal excludes diagrams or obsolete instructions, state that in both the scope and manifest so the recipient does not assume a complete product history.
Manual archive review sheet
Edit this outline, then copy or download it. It stays in this page and is not submitted or saved by this tool. Keep confidential records and personal information out. Save your work before leaving or reloading.
You can select and copy the text manually. This is a preparation worksheet, not a legal agreement.