How to proofread a long document

• Memos and reports where terminology drifts between sections written weeks apart.
• Executive summaries that contradict body detail because edits landed out of order.
• Appendices and tables that are correct in isolation but inconsistent with the narrative.

• Reading from page one to end in a single sitting—attention drops; errors cluster in the “boring” middle.
• Fixing sentences locally without updating dependent claims elsewhere.
• Letting formatting churn obscure real edits during team review.

Pass A — cold mechanics: spelling, grammar, numbering, captions, units, dates, xref labels. Work in short chunks (chapter, argument block, or time-boxed slices).

Pass B — warm coherence: narrative arc, tone, stakeholder language, and “does this still claim what we intend after Pass A?”

Keep a living terminology sheet (product names, banned phrases, capitalizations). One source of truth beats memory.

DraftLens reviews the manuscript in blocks with explicit budgets so models focus where the pipeline routes them—useful for long DOCX or PDF work where you want a second mechanical and consistency sweep. See multi-model review and the long-document use case for workflow placement.

• Organizational politics, approval chains, and “who owns this paragraph.”
• Fact-finding beyond what is in the manuscript and attached evidence (when enabled).
• Final voice polish—schedule human time for Pass B even if Pass A was assisted.

• Defined terms and product names consistent across body, summary, and appendices.
• Figures and tables referenced in text still match after late edits.
• Any AI-surfaced issue has a human disposition (accept / edit / reject with rationale).