The PPP fraud pipeline worked because the SBA released everything. Medicare's public data is fragmented, de-identified, and missing the features detection needs. Here's what exists on GitHub, where it falls short, and what CMS would need to release to let outside analysts do for healthcare fraud what one Python repo did for PPP.
The previous post described a Medicare fraud backtest nobody had built. I built it. 289 excluded providers across 41 states, matched to pre-exclusion billing data, compared against 3.39 million peers. Thirteen of fifteen features showed statistically significant differences — and the behavioral fingerprint is consistent enough to predict fraud in providers who were never excluded.
A walkthrough of building a Medicare fraud backtest overnight in Claude Code — from a plain-English spec to 289 matched providers across 41 states, a predictive model with AUC 0.79, and out-of-sample validation. Including the three times the pipeline failed, the data duplication bug, and the engineering decisions that shaped the final design.
