Inside a document: the extraction receipt audit trail

Every document you send Verinode, an invoice you forward, a certificate of insurance you upload, a job export you paste in, gets read by an extraction agent the moment it lands. That agent does not…

9 min read·Updated July 13, 2026
On this page

What this is

Every document you send Verinode, an invoice you forward, a certificate of insurance you upload, a job export you paste in, gets read by an extraction agent the moment it lands. That agent does not just guess at a few numbers and move on. It produces a structured record of its own work: which fields it found, which ones it turned into a saved record on your account, which ones it could not use, and a single completeness score summarizing all of it. Verinode calls that record the extraction receipt, and it is attached to every document that has gone through the pipeline.

This article is about reading that receipt: where to open it from a document row, what the completeness badge's color is actually telling you, what "we captured" means, and the two things you can do when a document did not come through cleanly. For the wider Vault page, its hero numbers, and the other three tabs (Tracking, Lists, Notifications), see The Vault: your Documents section and Reading the Vault hero. This article is the deep dive on a single document's own audit trail.

Verinode does not create records for you to fill in. A document's receipt only exists because that document actually flowed in, through email forwarding, an upload, a photo, a paste, or a voice note, and Verinode read it. Nothing here is invented on your behalf: you always decide what gets corrected, re-run, or removed.

Where to find it

  1. 1Open Vault from the sidebar, under My Data. The route is /data (bookmarks to the old /documents path still redirect here).
  2. 2Click any Explore tile or Recent uploads tile to open the four-tab detail view, or use the direct Uploads link if you are already inside Vault.
  3. 3Land on the Uploads tab: the full table of every document Verinode has processed for you, most recent first.
  4. 4Find the Status column. If a document has a receipt, a small colored percentage badge sits next to its Processed / Pending / Failed dot, for example a green 92%.
  5. 5Click that badge. The row expands underneath itself with a plain-language summary of what Verinode read.

Note

The badge only appears on documents that have a receipt. Very old documents captured before receipts existed have no badge at all, just the status dot. There is nothing to expand on those rows.

The completeness badge is the entry point

Hovering the badge (before you click it) shows one of three tooltips, depending on the score:

  • 80% or higher: "We captured most of this document. Click to review."
  • 40% to 79%: "Some fields are missing. Click to review or correct."
  • Below 40%: "We struggled to read this document. Click to correct it."

The badge itself is colored to match: green at 80 and above, amber from 40 to 79, red below 40. These are the same thresholds used everywhere completeness shows up on the platform, the Vault hero pill, the Tracking tab, so a red badge here means the same thing it means anywhere else: Verinode is telling you plainly that it did not get most of what this document type can contain.

Completeness is not the AI's confidence in its own answer. It is the share of the fields and records a document of this type could plausibly produce that actually turned into something saved on your account. A document can be read with total confidence and still score low on completeness, if the source document itself was missing information, or if what it contained did not map cleanly to anything Verinode's pipeline knows how to save yet. A document with nothing extractable in it at all (Verinode filed it but found no domain data to act on) still scores 100, because there was nothing expected of it in the first place; the score only drops when something was expected and did not make it through.

What "we captured" means

Click the badge and the row expands to a panel headed What we read from this document, with a Close link on the right. Underneath, a line reads:

We captured: followed by a comma-separated, plain-language list, for example "Total Amount, Period Start, Vendor Relationship, Line Items."

That list is a merge of two things the receipt tracks separately behind the scenes:

  • The document-level fields the extraction agent recognized in the text (an invoice's line items, a certificate's expiry date, a job export's records, and so on).
  • The capabilities that actually fired and wrote something to your account as a result (saving the vendor relationship, saving the line items, saving a certification record). A field being recognized does not automatically mean a save happened; the capability list is the part of the receipt that confirms it actually landed.

Verinode combines the two into one humanized list rather than showing you two overlapping, half-technical ones. If nothing meaningful came through, this line does not appear at all.

What you don't see spelled out, and why

The receipt itself, the underlying record Verinode keeps, also tracks capabilities that were skipped (nothing in the document matched what that capability needed), capabilities that outright failed (an error during the save), and any warnings the pipeline raised along the way (for example, a legacy file format that had to be handled differently). None of those three lists are printed row by row in the panel you see.

That is a deliberate choice, not an oversight. Naming the internal categories, "3 capabilities skipped," "1 warning", tells you nothing you can act on. Instead, the panel translates the same underlying signal into one plain instruction:

  • If anything failed outright, the panel 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 but the completeness score is under 80%, it reads instead: "Verinode wasn't fully confident on this one. You can correct the fields directly, or re-run extraction."
  • If the document scored 80% or higher with no failures, neither message appears, just the "We captured" line and the two action buttons below it.

What you can do about it

Two buttons sit at the bottom of the panel:

Edit fields. Opens the correction form inline, in place of the receipt panel. Six fields are editable: Document type (a dropdown covering Invoice, Renewal Notice, Certificate, Estimate, Email, Statement, Assignment, Payment Notice, Job Export, Carrier Scorecard, Incident Report, and Other), Total amount, Period start, Period end, Document date, and Reference number. Fill in what the document actually says, click Save corrections, and the row confirms "Corrections saved." a moment later. This is the fastest fix when Verinode read the document correctly but missed or misread one or two fields, you are telling it the right answer directly rather than asking it to try the whole document again.

Re-run extraction. Sends the document back through the same ingestion pipeline it went through the first time, working from the original file (or, for older documents, the stored text of what was originally sent). Use this after you have replaced the source file, for example if the version you first sent was a partial scan and you have since forwarded the complete one, or if a pipeline improvement since your last upload might read it better this time. On success the page reloads and you will see a fresh receipt, a new completeness score, and a new "We captured" list. If the retry request reaches the server but cannot complete, an alert shows the specific reason, or "Re-run failed" if none is given. If Verinode can no longer find the original file in storage (rare, and typically only on very old documents), the message reads: "Cannot retry: original file not found in storage. Upload the document again."

Tip

Reach for Edit fields first when the document itself was read correctly and only one or two values need a manual correction, it is instant. Reach for Re-run extraction when the source document has changed, or when several fields came through wrong at once, since a fresh pass through the pipeline is more likely to catch a pattern than one field correction at a time.

Needs Review takes priority over the receipt

A document can carry both an extraction receipt and an unresolved entity match at the same time, for example an invoice Verinode read cleanly but could not confidently tie to an existing vendor. In that case, a Needs Review chip ("Assign vendor," "Assign carrier," or "Assign job") sits in the Source column alongside the completeness badge in Status. Both point at the same row, but clicking either one opens the entity-assignment panel first, not the receipt. You have to confirm which vendor, carrier, or job the document belongs to before the row will expand to show you its extraction receipt instead. See Clients and carriers for how that assignment step works and why it matters for your benchmarks.

Checking against the source

Every row with a stored original file shows a small document-preview icon in the far-right action column, titled "View original document." Clicking it opens the actual file you sent, in a new tab, so you can put the receipt's numbers next to the real invoice or estimate side by side before deciding whether to edit or re-run. This is the fastest way to tell the two failure modes apart: a document that genuinely never had a due date printed on it, versus one where Verinode missed a due date that was sitting right there on the page.

Removing a document

The trash icon next to Edit in that same action column asks for confirmation before deleting anything: "Remove this document? This will also remove any signals sourced from it. This cannot be undone." A document's receipt, and everything it fed downstream, is genuinely gone once you confirm.

Heads up

Deleting a document erases its audit trail along with it, including any signals it fed into your benchmarks or findings. If a number looks wrong because of a bad extraction, correcting the fields or re-running extraction fixes the root cause while keeping the signal history intact. Save deletion for documents that should never have been sent in the first place.

Empty states

If your Uploads filters (search, bucket, or document type) match nothing, or you have not sent any documents yet, the table is replaced with a message describing how your document intelligence builds as files flow in, alongside a short list of what will show up here once your first documents land. If you have fewer than five documents on file in total, a prompt to connect email forwarding appears above the table, since that is the fastest way to keep the Uploads tab, and its receipts, filling on their own without repeated manual uploads.

Why this matters beyond the row

Completeness is not just a badge on one document. The score behind each receipt rolls up into the Tracking tab (an average completeness per vendor, client, or job) and into the Vault hero panel (an average across everything with a receipt). A document sitting at a low completeness score, or one still marked Needs Review, has not yet contributed everything it could to the signals behind your margin and benchmarks. Clearing a review, correcting a field, or re-running a stubborn document is the direct, concrete way to raise those numbers, and it is entirely in your hands: Verinode surfaces what it read and what it could not, you decide what to fix.

As an independent data trust, Verinode never sells your documents, or the values extracted from them, to carriers or anyone else. The receipt exists so you can verify, document by document, exactly what Verinode did with what you sent it, not to feed anything outward.

Data sources

Data sources

  1. 1.The document you sent (original file or stored text). Your business.
  2. 2.The extraction receipt Verinode generates when it processes that document. Verinode processing.
  3. 3.Corrections and re-run requests you submit. Your business.
Was this helpful?