Why it matters
Eleonora Drykina and Nina Žižakić describe retrieval failures caused by flattened tables and fragmented document context. Their approach retains section paths and chunk order, combines lexical and semantic retrieval, and expands context when needed. The talk also shows why model-judge verdicts need source checks. Private retrieval results do not independently establish answer accuracy.
My takeaway: Build held-out questions that require table headings, units, and evidence across sections. Preserve those relationships during ingestion and bound context expansion. Require precise source citations, inspect judge disagreements against the original document, and test that incomplete evidence produces an explicit limitation rather than an invented answer.