Reading the extraction receipt

What the extraction receipt is

Every document that lands in your Vault, an invoice, a certificate of insurance, a carrier scorecard, a job export, gets read by Verinode and turned into structured data: amounts, dates, vendor names, claim references, line items. The extraction receipt is the audit trail for that read. For each document it records what Verinode found, what it saved into your business record, what it could not save, and a completeness score that tells you how much of the document actually made it through.

This is not a confidence score. Verinode does not show you how sure the model felt about its own answer. The receipt shows you something more useful: of everything Verinode found in the document, how much of it turned into a real, saved record you can act on. A document can be read with total confidence and still score low here, if the fields it found did not match anything Verinode could save (a vendor invoice with no vendor name, for example). The receipt is about outcomes, not self-reported certainty.

As an independent data trust, Verinode does not quietly guess and move on when a document does not read cleanly. The receipt exists so nothing is a black box: you can see exactly what came in, what happened to it, and what to do next.

Where to find it

Open Data from the sidebar (/data). The Vault home page opens with three rows:

• The Vault hero panel at the top, showing your total document count and an overall Extraction Completeness pill.
• Explore, a row of tiles: Uploads, Tracking, Playlist, Reading, Saved, Notifications.
• Most recent, a row of tiles for your 8 most recently processed documents.

Click the Uploads tile, or any tile in Most recent, and the Uploads table opens. This table is where every individual document's extraction receipt lives.

The Uploads table

Each row is one processed document. The columns:

• Source, the vendor, client, carrier, or job the document is linked to, with the bucket it belongs to underneath (Vendor, Client, or Job). The name links out to that entity's page. If Verinode could not confidently match the document to an entity yet, an amber Assign vendor, Assign carrier, or Assign job badge sits next to the source, opening the assignment panel described below.
• Type, the document type Verinode classified it as (Invoice, Certificate, Estimate, Carrier Scorecard, Job Export, and so on), humanized from the raw classification.
• Amount, the total dollar amount extracted, right-aligned, or a dash if no amount was found.
• Processed, the date Verinode finished reading it.
• Signals, how many signals (patterns, anomalies, findings) this document contributed. A copper number links to Signals; a dash means none yet.
• Status, Processed, Pending, or Failed, alongside the completeness badge described below.

A checkbox column on the left lets you select rows. Select one or more and a floating bar appears at the bottom of the page: "N document(s) selected," with Remove selected (deletes the documents and any signals sourced from them) and Clear to deselect.

Three icon buttons sit at the right of each row: an eye icon to view the original document (opens the source file in a new tab, when one is on file), a pencil icon to edit extracted fields, and a trash icon to remove the document (with an inline confirmation: "Remove this document? This will also remove any signals sourced from it. This cannot be undone.").

Filtering. Above the table, search by vendor, carrier, job, or document type. Two dropdowns narrow the list further: bucket (All buckets, Vendors, Clients, Jobs, each with a live count) and document type (All types, plus every type present in your data, humanized).

The completeness badge and what it means

Next to Status, every processed document with a receipt shows a colored percentage badge:

• Emerald, 80% and above: Verinode captured most of this document cleanly.
• Amber, 40 to 79%: some fields are missing. Worth a look.
• Rose, below 40%: Verinode struggled to read this one.

Hovering the badge confirms which band you are in in plain language ("We captured most of this document," "Some fields are missing," or "We struggled to read this document"). Click the badge (or the row's Needs review badge, if one is present) to expand the row.

What the expanded receipt shows you

Clicking a document with no outstanding entity-assignment need expands the row into What we read from this document, a plain-English summary built from the receipt:

• We captured: a comma-separated list of every field and capability Verinode saved from this document (amounts, dates, reference numbers, line items, and whichever domain object fired, for example an invoice populating a vendor relationship, or a job export populating job records). Field names are humanized, never raw column names.
• If any part of the document failed to save cleanly, a line reads: "Some fields didn't come through cleanly. Edit the values yourself, or re-run if you've updated the source document."
• If nothing failed outright but the completeness score is under 80%, a line reads instead: "Verinode wasn't fully confident on this one. You can correct the fields directly, or re-run extraction."

Two buttons follow:

1. 1Edit fields opens an inline correction form: Document type (dropdown), Total amount, Period start, Period end, Document date, and Reference number. Change what needs fixing and click Save corrections. A confirmation ("Corrections saved.") shows briefly, then the page refreshes with your changes in place.
2. 2Re-run extraction asks Verinode to read the document again, useful after you have replaced or corrected the source file. It calls the retry pipeline directly and reloads the page on success.

Under the hood, the full receipt Verinode keeps per document also tracks skipped capabilities (parts of the pipeline that found nothing to do, because the document simply did not contain that kind of data) and structured warnings from the read. Those two lists are internal bookkeeping used to improve extraction quality over time; the operator-facing panel distills them into the two guidance lines above rather than surfacing raw internal labels, because "skipped: runVendorIntelligence" tells you nothing actionable. What you always get, in every case, is the field list, the completeness score, and a clear next step.

Tracking: completeness rolled up by entity

The Tracking tile opens a second view of the same underlying data, rolled up per vendor, client, or job rather than per document. For each entity tracked, the table shows:

• Docs, how many documents have been processed for that entity.
• Completeness, the average of that entity's document-level completeness scores, shown as the same colored pill (or "no receipt" if no scored documents exist yet).
• Fields, the count of distinct fields Verinode has ever captured for that entity, across every document.
• Signals, how many signals that entity's documents have contributed.
• Last update, how recently a document last touched that entity.

Filter by entity type (All entities, Vendors, Clients, Jobs) or by completeness band (High, 80% and above, Medium, 40 to 79%, Low, under 40%, or No receipt). Click any row to expand it and see the actual field list: "Fields Verinode has captured (N)," rendered as a chip for every field, humanized. If no receipt exists yet for that entity, it reads: "No extraction receipt yet, fields appear as documents are processed."

This is the view to use when you want to know not "how did this one document do" but "how much does Verinode actually know about this vendor, this client, this job, and how solid is it."

The hero panel's overall completeness

At the top of the Vault home page, the hero panel shows What Verinode Is Tracking: your total document count as the headline number, and, once you have documents on file, an Extraction Completeness pill averaging every document's score. The pill uses the same three-color scale as the per-document badge. Below it, a line adapts to your current state:

• If documents need review, it flags how many: "N document(s) need a quick review to unlock more signals."
• Otherwise, if documents are still processing, it says so: "N document(s) still processing, signals fire as soon as they land."
• Once everything is caught up, it reads: "Feeds every section of your business, vendors, clients, jobs, equipment, team, certifications, and more."

A trust line sits at the foot of the hero panel confirming how your data is protected at rest. This is the same honest posture that runs through the whole Vault: Verinode tells you plainly what it can and cannot see, and never oversells confidence it does not have.

Empty states

• No documents processed yet, or a filter matches nothing: the Uploads table shows an ambient empty state, titled "Your document intelligence builds as files flow in," previewing what is coming: automatic classification, entity matching, data extraction, signal detection, and a full audit trail.
• No documents at all before any have flowed in: the Recent uploads row on the Vault home page reads, "Recent captures appear here as documents flow in. Forward an email, upload a file, or paste a job export, the latest 8 land in this row."
• Tracking with nothing rolled up yet: "Tracking fills as documents flow in and Verinode extracts fields," with a note that each invoice, certificate, and statement is parsed by capability, captured fields stack up per vendor, client, or job, and clicking any row shows exactly which fields are tracked.
• Tracking filtered to nothing: "No entities match these filters," with a nudge to clear the search or switch the completeness filter.
• No cash balance, no data connected at all: the Vault hero subtext reads, "Connect a channel, email, uploads, voice, or photo, and your vault starts filling."

None of these are broken screens. They mean the underlying documents have not flowed in yet, or your filters are too narrow. As soon as a document is forwarded, uploaded, or ingested, its row and its receipt appear.

Best-practice example

A carrier scorecard comes in by email forwarding and lands in Uploads within a few minutes, showing Processed with a 55% completeness badge in amber. Clicking the badge expands "What we read from this document": Verinode captured the carrier name, the scorecard period, and two rating fields, but the guidance line reads, "Verinode wasn't fully confident on this one. You can correct the fields directly, or re-run extraction." Opening the original document with the eye icon confirms two fields were handwritten and hard to read. Rather than re-running (which would produce the same result from the same scan), click Edit fields, correct the two values directly, and save. The document's completeness reflects the manual correction, and its fields now show up cleanly under that carrier's row in Tracking.

Related reading

• Connecting your data
• Forwarding documents
• Clients and carriers
• The decision workspace

Data sources