See what stands between a document and an approved action.
Follow a supplier invoice from source documents to a prepared record. Introduce a missing reference or duplicate, inspect the checks, and see why the workflow waits for review.
All companies, documents, and amounts are fictional. The example runs entirely in your browser: nothing is uploaded, and nothing is written to another system.
Interactive example
One invoice. A visible decision path.
Fixed rules, not an AI model, read the labeled lines below, so the same document always produces the same fields.
01Source documents
02Extract fields
03Validate sources
04Human review
05Approved simulation
Each scenario is a separate exercise. Changing a source clears the extraction, review, and approval. Nothing is saved, and reloading starts over. Scenario contents and extracted values are not sent to analytics or attached to an inquiry.
Invoice input
Pinewell Service Co. · invoice INV-204
v1
invoice-204@v1 · Active source
01 SYNTHETIC INVOICE: fictional demonstration
02 Customer: Meridian Field Services
03 Supplier ID: SUP-014
04 Invoice ID: INV-204
05 Purchase order:
06 Service reference: SR-118
07 Description: Scheduled sensor inspection
08 Currency: USD
09 Quantity: 12
10 Unit price: 125.00
11 Invoice total: 1500.00
12 Supplier: Pinewell Service Co.
Reference source
Approved fictional service record · SR-118
v1
service-118@v1 · Active source
01 SYNTHETIC SERVICE RECORD: fictional demonstration
02 Service reference: SR-118
03 Purchase order: PO-074
04 Supplier ID: SUP-014
05 Currency: USD
06 Authorized quantity: 12
07 Authorized unit price: 125.00
08 Completion: Complete
09 Review: Approved in the fictional source system
The invoice has no purchase-order reference. The rules do not guess one from the service record; the supplier sends a corrected invoice instead.
Run the extraction and checks on the current source versions. Any earlier review or approval no longer applies.
02
Fields with source references
Run the extraction to see eight labeled fields and the source line for each. The source documents stay visible above.
03
Checks that control the next step
The checks cover required fields, duplicate invoices, source references, quantity × rate, currency, and service completion. Review opens only when all of them pass.
04 · Human review
Approve the prepared record, with its evidence.
The reviewer checks the invoice, the service record, any excluded copy, and the total. Here a checkbox stands in for that sign-off; it does not identify the reviewer or approve anything outside this page.
Review opens after the extraction runs and every check passes.
05 · Approved simulation
No approved record yet.
A record appears here only after every check passes and someone reviews the current sources.
Source, extraction, validation, review, and an approved draft.
From the example to a real workflow
The controls matter before the technology choice.
Know where each field came from.
The rules here read one exact, labeled format and record the source line and version of each field. Real documents vary, so a real system may need ordinary parsing, OCR, or an AI model, tested on representative documents against an agreed error policy.
Keep checks outside the extraction.
A value that looks right still needs an eligible source, a correct amount, a duplicate check, and a path for exceptions. Here, 12 units × $125 equals $1,500, calculated in whole cents; the revised source uses 12 × $130 = $1,560. Neither amount can skip review.
Tie each approval to the current evidence.
A changed invoice, service record, or duplicate decision clears the earlier approval. In a real system, the application also handles reviewer identity, access controls, audit records, and permission to write to other systems.
The JSON download contains the fictional sources, the selected scenario, the check results, and any simulated approval. The example shows how the controls behave; it does not measure extraction accuracy, savings, or production readiness.