Gate 1: heading hierarchy audit
Check that H1 appears once, levels are not skipped (H2 after H1, H3 after H2), and no heading is empty. Flat or inverted hierarchies break chunkers that use headings as split boundaries.
Gate 2: table and list integrity
Verify that tables have consistent column counts across rows and that lists are not broken by stray paragraphs. Broken tables and fragmented lists are the most common source of retrieval failures downstream.
Gate 3: section completeness
Every heading should have at least one paragraph below it. Headings without content create empty chunks that waste embedding space and confuse retrieval ranking.