Publicly funded healthcare data

Healthcare's public data, in everyone's hands.

The law already requires health plans and hospitals to publish this. It does not require a format, does not require the arithmetic to check out, and almost nobody reads it. Project Crossfoot collects it, validates it against itself, and puts it within reach — no middleman.

01 · The data
2.1 billion
Lines of hospital pricing data processed
Healthcare prices
20,125,680
hospital price line-items, pulled from the machine-readable files hospitals are required to publish and checked for internal consistency
5,381hospitals 529KMedicare rows 3,152counties 93Kbilling codes
Prior authorization
requests decided in CY2025 — standard and expedited, summed from the counts the plans themselves published under CMS-0057-F
658filings 124insurers 48states 7.8Mdenied
02 · Healthcare costs

Hospitals bill $5.03 for every dollar that actually gets paid

Every hospital publishes what it bills Medicare, and most now publish their own price lists too. Put side by side, the numbers tell two stories: the bill and the payment are different worlds, and the same care carries wildly different prices depending on the building it happens in — and 342 hospitals publish price files whose own numbers contradict each other.

One procedure, every hospital: SEPTICEMIA OR SEVERE SEPSIS WITHOUT MV >96 HOURS WITH MCC
the most common Medicare hospital stay (MS-DRG 871) — no maternity here, Medicare is 65+ — what 2,661 hospitals billed for it, on average
$26,595the cheapest twentieth bill less $65,249half bill less, half bill more $173,762the most expensive twentieth bill more

Same procedure, same payer — and the high end bills 6.5× the low end. That spread, item by item and hospital by hospital, is what the price section maps.

03 · Prior authorization

The no that nobody checks

Before treatment, the insurer decides whether to allow it. Health plans now have to publish how often they say no — 124 insurers' filings, collected and checked here. Follow the requests down: most denials are never challenged, and when someone does push back, the insurer's own review often changes the answer. And 28 filings publish numbers that contradict their own arithmetic.

requests decided 73,167,403 every decision the filings put a number on, CY2025
denied 7,752,713 10.6% of requests — about 1 in 9
denials appealed 3,560,291 64% of denials, where plans report appeals — the rest were never challenged
denials overturned 130,906 4% of appeals succeed — the no becomes a yes

Appeal counts cover standard denials, from the 365 filings that publish them. Every number is summed from what plans published; a plan that published no count is left out rather than guessed at.

04 · The analysis

Which hospitals charge more than their services explain

An MRI costs more than a blood draw everywhere — and some hospitals charge more for everything. So we compared 12.1M prices that 1,202 hospitals published for the same services, and gave each hospital one score: ×1.00 means it charges the going rate for what it does, ×2.00 means double the going rate. Most land close to ×1.00. The outliers are the story.

Every scored hospital, on one line
1,202 hospitals by their score — the pile in the middle prices close to the going rate; the orange tail does not
×1.00 — the going ratehalf of all hospitals land in this shaded band×2 — double the going rate →Boston Children's Hospital — ×4.09×0.5×1×2×4each bar counts hospitals · cheaper than expected ← → dearer than expected
×0.76–×1.30
where half of all hospitals land — from 24% below the going rate to 30% above the going rate
74
hospitals charge at least double the going rate for the same services
×4.09
the most expensive: Boston Children's Hospital, MA — about 4× the going rate
05 · Open source

Built in the open

This cannot be overstated: the dataset and the pipeline that produced it are given away. Every number is free to download, the crawlers that collected the documents are public code, the AI-agent workflow that read them is documented, and the calculations and the models — county outcomes and the fair-price index alike — are open source. Check the work, rerun it, or point it at data we haven't touched.