Document Extraction Reconciler

How to use Document Extraction Reconciler

An extraction can have the right total and still assign amounts to the wrong rows. Reconcile independent reference rows with observed extraction rows before accepting a document-processing workflow.

1. Prepare expected rows by checking the original through your own review process.
2. Paste observed key/amount rows using the same units and sign convention.
3. Resolve missing/duplicate keys and row mismatches; do not rely on totals alone.

Good to know

• This does not read a PDF or certify the original reference. Only entered key/amount records are reconciled; currencies, units, tax and document semantics are not inferred.
• Signed plain decimal amounts are supported. Duplicate keys contribute to totals but are excluded from unique-key comparisons; matching totals can conceal individual errors.
• Use the exact displayed CSV header. Quoted commas, doubled quotes and quoted line endings are supported; malformed quoting or uneven rows are refused. CSV exports prefix text cells resembling spreadsheet formulas with an apostrophe.
• Local deterministic processing: no AI model calls, uploads, account connections, messages or saved input. Review any downloaded artifact before sharing.