Skip to main content

Document Processing

After upload, every document goes through a fixed sequence: automatic extraction first, then explicit human review, then a registration decision. Nothing enters the Registry without that decision. This is the review-first rule that shapes the rest of the product.

Use this page as a step-by-step guide to the lifecycle of a single document.

Processing pipeline​

  1. File is discovered — through upload or via a bound source folder.
  2. Parser extracts key fields — counterparty, totals, VAT numbers, dates, line-level data, and other metadata.
  3. Document enters review state and appears in the preview pane.
  4. You review the extracted data and choose what happens next.

You stay in control of what is accepted into the Registry. The parser is a draft, not a decision.

Reparse options​

The preview offers two reparse paths:

  • Reparse (Canonical Route) — the stable production extractor, optimised for predictable results.
  • Reparse (ZenoLab) — an experimental extractor, available where enabled, used to test improved parsers on real documents.

Reparse when:

  • extraction quality is visibly low,
  • the source file has been updated or replaced,
  • you want to compare canonical and lab behaviour on the same document.

Decisions you can take after parsing​

When the preview is ready, the right-side actions let you:

  • Register — accept the document into the Registry. You may need to pick a period before confirmation.
  • Mark Reviewed — record that you checked it without registering it (useful for documents that should stay out of the ledger).
  • Exclude — explicitly remove it from further consideration. Use this for duplicates, spam, or pages that do not belong.
  • Send to Accuracy Review — report structured extraction issues without sharing raw bytes (see below).

Only registered documents appear in Registry-backed ledgers, matching flows, and exports.

Accuracy review flow​

Send to Accuracy Review opens a checklist modal. You:

  1. Select fields that are mismatched compared with the original document.
  2. Add free-text context where the UI invites it.
  3. Submit the checklist.

What is shared with the system:

  • the keys of the fields you flagged,
  • anonymised structural signals,
  • a virtual snapshot of the source when bytes are available locally.

What is not shared: raw, unredacted document content as a matter of normal operation. The accuracy-review flow exists specifically so we can improve parsing quality without seeing your documents end to end.

Opening the original source​

The Open source action tries to open the original file referenced by the preview.

Notes that often come up:

  • In local-folder session mode, browser security may block direct folder opening for security reasons.
  • Safari in particular does not allow opening local folder paths from a web UI.
  • If the source reference cannot be resolved, the UI shows Source link unavailable.

When local context goes stale, Change folder reconnects it cleanly.

What stays manual on purpose​

By design, a few things never happen automatically:

  • Field-level correctness — you decide if a number is right.
  • The final register / mark-reviewed / exclude outcome.
  • Period selection before registration when periods are configured.

These are the moments where Zeno CY is slower than a hands-off automation tool — and where it earns the audit trail you can defend.