Example analysis on synthetic data — not a customer. The ten payments below are a test fixture that lives in this product’s repository. They are not a real business and not a real ledger.
Start at the total. Finish at the line in the file.
Three clicks, no account. The figure at the top breaks into categories, a category breaks into findings, and a finding breaks into the payments it was built from — each one carrying the name of the file and the line it sat on.
Source: backend/tests/test_financial_golden_datasets.py · DUPLICATES · TestDuplicateDetection (6 tests)
Level 1 · How much money is involved
654.30
across 2 findings, from 10 payments read
This file declares no currency, so the figure is shown exactly as the checks report it, with no symbol. Amounts are never converted between currencies, and a file that mixes them is reported as mixed rather than added up.
Nothing below is hidden behind a sign-up. Press the button and keep going until you are looking at a line number.
The file that went in, all ten lines
Not a sample of the file — the file. Ten payments is small enough to print in full, and printing it in full is the only way to show that the two findings above were not picked out of a much larger set that also contained things we would rather you did not see. Negative amounts are money leaving the business.
| line | date | vendor | amount |
|---|---|---|---|
| 2 | 2026-01-19 | Acme Supplies | -212.40 |
| 3 | 2026-01-28 | Acme Supplies | -212.40 |
| 4 | 2026-01-15 | CloudHost | -180.00 |
| 5 | 2026-02-15 | CloudHost | -180.00 |
| 6 | 2026-03-15 | CloudHost | -180.00 |
| 7 | 2026-01-21 | Initech | -212.40 |
| 8 | 2026-02-03 | Fabrikam | 900.00 |
| 9 | 2026-02-10 | Fabrikam | -900.00 |
| 10 | 2026-03-04 | Umbrella Logistics | -441.90 |
| 11 | 2026-03-09 | Umbrella Logistics | -441.95 |
Open a finding above and the lines it was built from are highlighted here.
Example analysis on synthetic data — not a customer. No customer data appears anywhere on this site.
What the checks looked at and chose not to flag
Eight of the ten lines above were not returned, and three of them were near misses that a careless tool would have reported. This is the half of the result that decides whether a report like this is worth opening a second time.
Three identical hosting payments, one a month
Regular monthly payments are a subscription, not a duplicate.
test_a_monthly_subscription_is_not_called_a_duplicate
An invoice and its refund, same supplier, same amount
Money out and money back in are opposite directions, not a repeat payment.
test_a_refund_is_not_a_duplicate_of_its_invoice
The same amount paid to two different suppliers
Two suppliers charging 212.40 is a coincidence, not a double payment.
test_the_same_amount_to_a_different_vendor_is_not_a_duplicate
Each line names the test that pins the behaviour. If someone changed the checks so that a monthly subscription started coming back as a duplicate, that test would fail before the change reached anybody.
What was run over the file
- identical payments to the same supplier, in the same direction
- payments within 5% of each other to the same supplier
- many small payments to one supplier
Three checks, written in Python, producing the same answer every time they see the same file. No language model decides what is flagged, and no figure on this page was written by one.
Example analysis on synthetic data — not a customer. No customer data appears anywhere on this site.
Now run it on a file that matters.
Or walk through a larger example first, on a 51-line construction ledger, without uploading anything.
A finding is something worth a human look, not a conclusion and not accounting advice — you decide what each one means.