From a public SAR narrative to a synthetic case
Regulators, standard-setters and financial intelligence units publish a steady stream of anonymised case narratives and typology guidance. A write-up describes how funds moved, which controls failed to catch it, and what the filing institution noticed first. Names and identifiers are stripped, but the shape of the behaviour survives: the sequencing, the amounts relative to the customer profile, the misuse of an otherwise ordinary product.
Those publications are our starting point, and we use them for their patterns rather than their text. A narrative gives us a structure to work from: the roles involved, the documents that should exist, the points where an alert would plausibly fire. No sentence of the source is carried over. What transfers is an understanding of how that kind of scheme unfolds and where it leaves traces.
Everything else is invented. The customers, the accounts, the transactions, the screening hits, the policies the agent works against: all fictional, built to be internally consistent rather than copied from any real file. A case needs enough ordinary activity around the suspicious core that the agent must actually discriminate, and that surrounding life is designed with the same care as the typology itself.
Each case keeps provenance for the pattern it draws on, held separate from what is invented. The record states which published guidance informed the design, so a reviewer can check that the behaviour in the case matches the behaviour the guidance describes. That separation matters in both directions: it keeps the case honest about its roots, and it keeps the source material from leaking into the case text.
Evidence can also arrive in stages. Real files rarely land complete; the analyst requests bank statements, asks the relationship manager for context, or waits on an information request before the picture settles. Cases reproduce that rhythm where it fits the typology, so the agent must judge when the file supports a call and when the right move is to ask for more. Deciding under uncertainty is part of what the benchmarks grade.
Before any case enters a benchmark, it will go through review. A reviewer will check that the intended decision is reachable from the material provided, that the evidence supports it without depending on a trick of wording, and that the difficulty sits where it was designed to sit. Cases that fail go back for revision or stay out of the sealed set.
The result is a caseload with clear lineage and no borrowed text. No source narrative is reproduced, and no real customer data is used anywhere in the process. What the agent sees is fiction built on published patterns.