Find the pages your RAG should reindex.
Audit embedded-image OCR, content freshness, and chunk creation before treating retrieval misses as a vector-index problem.
Page ingestion manifest
strict 9-column CSV
synthetic sample
synthetic sample
Decision rule: reindex for missing image OCR, stale content hash, or zero chunks. Retrieval hits remain diagnostic context, never a standalone trigger.
Synthetic eight-page manifest loaded. Ready to audit.
Evidence-backed queue
page union
reasons preserved
reasons preserved
No queue yet.Run the manifest through the production parser to see exact page-level repair evidence.
REINDEX DECISION
—
IMAGE OCR—
HASH FRESH—
HIT PAGES—
CLEAN PAGES—
Boundary: This audit reads supplied metadata only. It does not inspect PDFs, perform OCR, query Qdrant, benchmark HNSW, validate retrieval quality, or access the source author’s system. Zero retrieval hits are evidence to investigate, not a reindex trigger by themselves.