Last updated: 2026-04-25

Why didn't my statement extract?

You uploaded a statement, the document landed in your vault, but the per-property entries table is empty (or shows a "needs review" banner). Walk this checklist.

Check 1: did the classifier tag it as a statement?

Open the document detail page. Look at the document type chip near the top.

• If it says Agency Statement ✓ — the classifier got it. Move on to check 2.
• If it says Other, Invoice, or anything else — the classifier didn't recognise the format. The most common cause is a one-property statement that looks more like an invoice. Manually re-tag the document type to statement and reprocess.

Check 2: was the file's text layer readable?

PDFs come in two flavours: text-layer (text is digitally present and selectable) and image-only (a scan of a printed sheet). Anchorlet's deterministic XLSX parser only runs on XLSX. The PDF path uses an LLM that can read both, but image-only PDFs are slower and less accurate.

If your statement was a scan:

• Open the document detail page and click "View source" to confirm the text is readable.
• If you can re-export from the agency's portal as a text-layer PDF or as XLSX, do that instead.

Check 3: is the agency format unusual?

The deterministic XLSX parser is tuned for Ray Cooke Auctioneers. Other agencies fall through to the LLM path automatically. If your file is XLSX from an agency the deterministic parser doesn't know, it'll throw a parse error and the document lands in your vault without entries — but the LLM path doesn't run on XLSX automatically.

Workaround: re-export the statement as a PDF and upload that. The LLM extractor handles arbitrary layouts.

Check 4: reconciliation mismatch

If you see a yellow Reconciliation mismatch banner on the statement detail, the rows + totals don't agree. Anchorlet still ingested everything; it's just flagging that the agency's spreadsheet has a manual override or a row outside the standard shape. Check the raw_extract metadata on the document — usually a deposit, a credit note, or a one-off line that doesn't fit the standard rent/fee/expense grid.

Check 5: still nothing — reprocess

If the file is right and the classifier tagged it correctly but you still don't see entries, hit the … menu on the statement detail and click Reprocess. The old statement gets soft-deleted (recoverable from Recently Deleted for 30 days) and the source file is re-extracted from scratch.

If reprocess produces the same empty result, the file genuinely lacks the structured fields the extractor looks for. The next step is to upload it as a regular document (so the PDF stays accessible) and enter the totals manually for now. We're tightening the LLM extractor's tolerance for non-standard formats — a corrected version will land soon.

Still stuck?

Contact support with the statement's id (in the URL) and we'll dig in.