About Fraudhound
Fraudhound is an independent watchdog that uses statistical anomaly detection to scrutinize federal contracting. Every night, we score every contractor in the federal contracting database against six statistical detectors. The 50 most anomalous get profiled in a daily briefing.
We are not the FBI. We do not allege fraud. We surface statistical patterns that can indicate fraud, scope creep, non-competitive procurement, or contract gaming — and we show our work so you can judge for yourself.
How we detect
Six statistical tests run every night against the full federal contracting database.
Benford's Law
Looks for unnatural patterns in the first digit of award amounts. Real-world dollar amounts follow a predictable distribution where the digit "1" appears in ~30% of values. Deviations can indicate manipulation or manufactured numbers.
New Entity Sole-Source
Brand-new contractors winning large non-competitive awards as one of their first contracts. Fits the pattern of entities created to capture specific awards.
Modification Growth
Detects parent contracts whose modifications grew much faster than peer contracts in the same NAICS code. Extreme growth can mean scope creep, inadequate competition, or post-award gaming.
Isolation Forest (Multivariate Outlier)
Machine learning method that finds contractors whose overall profile — agency mix, sole-source rate, NAICS spread, modification frequency — is statistically unlike anyone else's.
Sole-Source Concentration
Contractors that win far more non-competitive awards than industry peers. Sole-source contracts are legal and sometimes necessary; extreme concentration is unusual.
Award Velocity
Contractors whose recent award count is statistically far above their own historical baseline. Could indicate sudden favoritism or a relationship shift at an agency.
The briefs
For the top 50 contractors each day, we generate a short three-paragraph brief using Claude. The model receives only the structured detector outputs, so the same inputs always produce the same brief. Every brief is auditable and reproducible from public data.
Data sources
- Contract data
- USAspending.gov (public API + archives)
- Coverage
- Department of Defense + Health and Human Services, FY2024–2026
- Refresh
- Nightly, ~2am ET
What we don't claim
- ·A high composite score is not proof of fraud.
- ·Detectors are statistical signals, not legal evidence.
- ·Many flagged contractors are large established firms whose anomalies have legitimate explanations.
- ·Investigation, audit, or document review is required before drawing conclusions.
Who built this
Fraudhound was built by Kamron Arabi as an independent accountability project. Source code is open on GitHub.
Contact: [email protected]