Design the failure. Prove your product catches it.
You ship software that reads money. The only honest test is: plant a known mistake, then check whether your product finds it.
The problem
You can't QA an anomaly detector without anomalies you control. Production data has unknown ground truth — you never really know what you missed. Synthetic data without planted, labelled mistakes proves nothing.
What Fictix gives you
Describe the failure in plain language — duplicate bill, ghost employee, wrong revenue period, books that don't reconcile — and Fictix plants it and remembers exactly what it planted and where. Run your product against the company, and Fictix scores it: precision and recall against known ground truth, scenario by scenario.
Regression you can trust
Curate a test matrix of scenarios that must pass before release. Track results over time. File tickets that point at the exact transaction, because you planted it — no more “we think it missed something.”
Questions
Do I need to write JSON to design a scenario?
No. You describe the anomaly in plain language; Fictix injects it into the company's books and records it as ground truth.
How do I know what my product missed?
Fictix knows what it planted, so it scores your product's findings on precision and recall per scenario — the misses are explicit.
Can scenarios gate a release?
Yes. Group scenarios into a test matrix that must pass before shipping, and track regression over time.