🕵️ The Perfect Audit Crime Challenge
Synthetic general ledgers with unlabeled planted fraud. Flag the fraudulent journal entries; the scorer ranks you on the held-out ground truth.
Two tracks, same ledgers:
- Track A — ledger only. Structured, relational and memory frauds are catchable. The mimetic perfect crime — fraud drawn from the ledger's own normal distribution — provably is not.
- Track B — ledger + evidence. You also get the ISA-520 expectations and ISA-505 evidence anchors. Now the aggregate-inflation and fabricated-counterparty crimes become catchable.
The gap between the two tracks on the mimetic family is the point: it maps where ledger analytics ends and external evidence must begin.
Research benchmark on purely synthetic data. A high score is not a deployable detector, and this is not a how-to for fraud — the un-catchable items are why assurance needs external evidence.
Submission: a CSV/JSON listing the document_ids you flag as fraudulent (optional score column for
ranking). Metric: PR-AUC on the held-out labels (plus precision / recall / F1).