About

In February 2025, the HHS DOGE team open-sourced the largest Medicaid dataset in department history: 227 million claims, covering $1.09 trillion in spending from January 2018 through December 2024. For the first time, complete provider-level spending data is available to every citizen.

MedicaidWatch was built to make this data accessible. Raw data alone doesn't tell stories. We built five detection algorithms that scan every provider in the dataset, looking for billing patterns that warrant investigation.

We believe in transparency. Every finding on this site includes the exact numbers, methodology, and peer comparisons used to identify it. Nothing is hidden. Readers can verify every claim against the public dataset.

Our Principles

  1. We don't accuse — we illuminate. Anomalies are patterns that warrant investigation, not evidence of wrongdoing.
  2. Context matters. Every flagged provider includes an explanation of why the pattern could be legitimate.
  3. Show the math. Every number can be verified. Every methodology is documented.
  4. AI as a tool, not an authority. AI generates narratives from data — humans must verify and act.

Technology

MedicaidWatch is built with Python and DuckDB for analytics, Astro for the web framework, and Claude for narrative generation. It deploys to Cloudflare's edge network. The entire analysis pipeline runs on commodity hardware.

Open Data

The underlying dataset is publicly available from HHS Open Data. Provider identity data comes from the NPPES NPI Registry. Exclusion data comes from the OIG LEIE. All sources are public and free.