[ 00 / MANIFESTO ]

Fraud detection is an engineering problem.

For fifty years, forensic accounting research has produced models that work: Benford’s Law deviations, discretionary accruals, accrual quality scores, structural distance-to-default. The literature is public. The mathematics is sound. And almost nobody runs it — not systematically, not across entire markets, not every reporting period.

Why? Because running forensic mathematics at market scale was never a research problem. It’s a data engineering problem, a calibration problem, an infrastructure problem. Academia doesn’t build pipelines. Funds build them for their own book and tell no one. Sell-side has no incentive to find fraud at all.

$ read --manifesto →

Detection tools

Regulatory regimes

Verdicts in the engine

[ 01 / THE FOUNDATIONS ]

/methodologyThe fraud literature works. Almost nobody runs it.Ask a quant about alpha factors and you'll get a story about decay: published anomalies stop working once the paper is out, arbitraged away by everyon…methodology · 2 min read /methodologyThe engine produces clues, never verdicts.Early in building BearSignal, we made a mistake that took weeks to undo. It felt natural: if three detection models flag the same company, surely the …methodology · 2 min read /engineeringTwo regulators, one schema: building fundamentals pipelines for the U.S. and Hong KongForensic mathematics has a boring dependency: it only works on clean, comparable, point-in-time fundamentals. Benford tests need raw reported digits. …engineering · 3 min read /engineeringNULL is a feature: the most boring rule we refuse to breakEvery data team has a rule like this somewhere in a style guide, half-enforced: don't fabricate missing values. Ours is absolute, and it has a name we…engineering · 2 min read /careersWork on something with an answer key.Most software ships and you never find out if it mattered. Our output gets graded — by enforcement actions, by restatements, by what happens to a stoc…careers · 3 min read