Rule review · demo
How your fraud rules perform, measured on your own data
WoWyrd measures how the fraud rules you already run perform on your own data, inside your own environment. It shows which rules to retire, merge, retune or add, with the effect in your own costs. It works from files your current rule engine can export: your transactions, your rule-hit and decision log, fraud labels and your unit costs. It changes nothing in your engine.
On this page: the reports for synthetic demo clients, then an example of your regulator's requirements for your licence.
Demo reports
We measure a bank's own fraud rules on its own files. The result is three reports: a short one for management (two to three pages), an extended one for management with every chart and check, and a detailed one for the rule team. Each report has a PDF version.
Why 3 demo clients? They are 3 different public datasets. Each synthetic dataset stands for one kind of client, so you can open the one closest to your business. Each row names its dataset, with a link and its licence. One more synthetic client gives its own rule file, so you can see how we check your rules themselves.
| Demo client | Management report (short) | Management report (extended) | Technical report |
|---|---|---|---|
| Bank transfers Stands for a bank that watches account-to-account transfers for money laundering. Dataset: IBM Transactions for Anti Money Laundering (AMLworld), HI-Small (IBM; Altman E. et al., 2023). Synthetic bank transfers with labelled money-laundering patterns. Licence: CDLA-Sharing-1.0. We converted it to the review's input format and took a sample of 58,331 transfers. | Open · PDF | Open · PDF | Open · PDF |
| SEPA transfers Stands for a European bank or payment institution whose customers are tricked into paying fraudsters (scams on SEPA credit transfers). Dataset: Synthetic SEPA Credit Transfer Fraud: APP scams, name checks and money mules (Gaydabura D., 2026). Synthetic SEPA credit transfers with labelled scams and money-mule accounts. Licence: CC BY 4.0. We converted it to the review's input format and took a sample of 59,137 transfers. | Open · PDF | Open · PDF | Open · PDF |
| Mobile money Stands for a mobile-money provider: payments, transfers and cash-out from e-wallets. Dataset: PaySim: Synthetic Financial Datasets For Fraud Detection (Lopez-Rojas E. et al., 2016). Synthetic mobile-money payments, transfers and cash-outs with labelled fraud. Licence: CC BY-SA 4.0. We converted it to the review's input format and took a sample of 54,510 transactions. | Open · PDF | Open · PDF | Open · PDF |
| One client, 3 business lines The same 3 datasets as business lines of one client (Bank transfers, SEPA transfers, Mobile money): a breakdown per line and a total. | Open · PDF | Open · PDF | Open · PDF |
| One client with its own rule file (synthetic) We replay your own rules on your history and show how often our replay agrees with your engine's log. For a rule that agrees, we show what a different threshold would have done on your history. One rule declines payments: there, a changed threshold can raise the false-alarm cost, because good payments that other rules still alert on move from your friction cost to analyst time. One rule is written "at least" where the engine says "more than": its agreement is below the floor, so it gets no what-if. Three rules are marked as AML rules: the reports count the payments just under each threshold that we picked for an analyst to review, and show the share an analyst marked suspicious in the sample of the report 35 days earlier. The reports also say what changed since that earlier report, and search each number threshold on the client's history for a better value, judged on the later part of the history it never saw. Data: Made-up payments from our own seeded generator, no public dataset and no real client. The rule file is a demo file; its consent line names no person. The earlier report is the same data up to 35 days before the last payment, and its analyst marks come from the synthetic fraud labels. For this demo we set the logged hits of the declining rule "Wire total from one sender in a week" to 'decline' so the log matches the rule file. Other rules that fired on the same payments now show them as declined too. Download an example rule file (the format to give us your own rules, with a note on every key) | Open · PDF | Open · PDF | Open · PDF |
Data that we derived from a dataset keeps that dataset's licence.
Your regulator's requirements for your licence
For a synthetic non-bank payment provider in Ukraine: the requirements that apply to its licence, the WoWyrd rule that answers each one (released, or still in development), and what changed in the last 90 days.
This is not legal advice. Your compliance team decides how a requirement applies to your business.
Regulatory watch
Your rules are mapped to current requirements. We flag when a new requirement affects them. We follow the official publications of regulators in Ukraine, the EU, the UK and the US. The National Bank of Ukraine text we map to is a draft, not adopted rules.
This is not legal advice. Your compliance team decides how a requirement applies to your business.