What We've Seen on Real Medical Bills

Last updated: 2026-04-27

Disclaimer: The numbers below describe what the analyzer's output looks like in aggregate. They are not medical or financial advice and do not characterize the legal or clinical accuracy of any specific bill.

How these numbers are produced

When a bill is analyzed, the file type and a few aggregate signals from the response text are incremented in process memory. The bill file, the response text, and any identifying detail are never stored. Code system mentions are detected by simple regex over the response (counting whether the AI named CPT, HCPCS, ICD-10, NDC, DRG, a modifier, or a revenue code), so these are heuristic counts, not a certified audit. Counts reset to zero on each deployment of the application, so totals below cover the period since the most recent deploy.

Total analyses

Awaiting first analyses since the most recent deploy. Stats will populate as the tool is used.

File type mix

No file type data yet for the current deploy window.

Code systems mentioned in analyses

The percentage of analyses where the AI response named each coding system at least once. Detected by regex over the response text, so treat these as approximate. For definitions of each code system, see the codes explained page.

No code mention data yet for the current deploy window.

How often issues were flagged

Awaiting first analyses to compute issue flag rate.

What is NOT stored

The bill file, the response text, dollar amounts, dates of service, provider names, patient identifiers, and IP addresses are never written to any store. Only the per-deploy in-memory counters above are kept, and they reset on the next deploy. For details on how uploads are handled end to end, see the methodology page.