Developer finds 52% duplicate nodes, 32% missing docs in knowledge graph

A systems administrator building a Rust-based knowledge graph for Dominican banking regulations discovered two silent failures: 52% of nodes were duplicates and 32% of source documents never entered the graph. The duplicate issue stemmed from re-running ingestion without uniqueness checks, while missing documents were silently dropped due to a 100-character text threshold that excluded scanned PDFs. An OCR fallback recovered most missing documents, but some corrupted native text remains unresolved.