Document Chunk Source Coverage

How to use Document Chunk Source Coverage

A retrieval workflow needs to preserve the relationship between chunks and original document elements. Check entered source IDs and sizes without uploading documents or assuming character counts are tokens.

1. Export minimal original element IDs, page/order and authorized character-count metadata.
2. Enter each chunk’s source-element sequence and the separator and maximum character assumptions.
3. Resolve missing or unknown sources, explain intentional overlaps and inspect actual retrieval quality separately.

Good to know

• Each chunk must refer to a single source document with contiguous membership positions and increasing source order. Missing source lengths remain unknown.
• At most 100 elements and memberships; maximum declared chunk size 1,000,000 characters and separator 100 characters. No PDF parsing, document content inspection, token counting or retrieval scoring occurs.
• Each input field is limited to 8,000 UTF-16 characters. CSV needs the exact displayed header and at most 100 data rows. Quoted commas, doubled quotes and quoted newlines are supported. Duplicate identities are refused unless the tool explicitly reports them as findings.
• Markdown and CSV are editable review artifacts, not certificates. Spreadsheet-like text is apostrophe-prefixed in CSV to reduce formula execution risk; check the actual importer before sharing. Decimal ratios are rounded to at most 12 places.
• Runs only on the records you enter, locally in your browser. No AI call, provider connection, account inspection, upload, automatic outreach or saved input. Use neutral identifiers rather than personal or secret content.