rareagent@work:~$
problems·news·[reports]·docs·start-here
|
services:pricing·industries·enterprise
|
trust·feedback
> Read full report
Rare Agent Work

Rare Agent Work · Open Data Edition

Rev 1.1 · Updated September 21, 2026

New
Free · open access12 minute brief + runnable open-source toolkitPatients facing scheduled procedures, journalists, benefits teams, and policy researchers

The Price Is Public. Nobody Can Read It.

What 10.1M hospital price rows reveal about American healthcare pricing

US hospitals are legally required to publish every price they charge — and bury it in files too large to open. This report reads them, and shows you how to check your own hospital in an afternoon.

What this report gives you

  • 01Hospital price files are readable by anyone: a streaming parser turns a gigabyte-scale CMS-mandated file into a queryable database in under a minute, on a laptop, for free
  • 02Sticker prices are routinely an order of magnitude above what the system actually transacts at — up to 77x on a CT scan within one hospital (case-rate artifacts filtered)
  • 03The same procedure in the same city varies 7–14x in negotiated rates between hospitals; a brain MRI spans $311 to $8,675 across five institutions
  • 04The unsolved problem is not reading the data (Turquoise, Serif, DoltHub already do) — it is an open, continuously refreshed, redistributable national dataset
  • 05Every number here is reproducible: parser, query CLI, and stats generator are open source, and the methodology excludes cross-setting and case-rate artifacts

The finding that changes your next decision

“The $17,058 sticker price for a CT scan with contrast sits 77x above the best negotiated rate at the same hospital — and until this data was mandated public, no patient could see either number. Transparency without tooling was functionally secrecy.”

Read the full report — free ↓
✓ All sections open · No sign-up · No paywall
✓Full preview before purchase✓Cited sources✓Updated September 21, 2026✓Human-authored✓No sign-up required

Why this report exists

Since 2021, every US hospital must publish a machine-readable file listing its sticker price, cash price, and every insurer-negotiated rate. Compliance is now broad — but the files are deliberately, or at least functionally, unreadable: gigabyte-scale JSON that crashes spreadsheets and browsers. The result is a transparency law that publishes everything and, without tooling, informs no one. We built a streaming parser that turns any hospital's file into a queryable database, and ran it across five hospitals in four states. The verified spreads are far larger than the 2-3x folklore: within one hospital, sticker prices run up to 77x the best per-code negotiated rate for the same procedure, same setting. Across hospitals, identical procedures differ 7-14x in negotiated rates. The entire US hospital price dataset — every hospital in the country — fits on a laptop. The problem is not data availability — Turquoise Health, Serif Health, DoltHub and PatientRightsAdvocate.org all read it at scale. The unsolved problem is maintenance: everything comprehensive is proprietary, and every open dataset we could find is stale.
🏛️

What's at stake

  • →Transparency without readability is not transparency. The binding constraint on the price transparency rule is not hospital compliance — it is file formats that make the data inaccessible to the people the law was written for.
  • →The cash price is not the floor. At one hospital the cash price for a knee replacement is $937 — far below most negotiated rates — while at another the cash price equals sticker. Patients who assume either direction will be wrong; the only way to know is to look it up.
  • →Negotiated-rate spreads within one hospital mean "your insurance is good here" is unknowable without plan-level detail — which the files do contain, per plan.
  • →For benefits teams: the same employer-funded procedure can cost 7x more depending on contracted hospital. Price-data-informed steering is now technically trivial; it was designed to be hard.
  • →For journalists and researchers: commercial readers (Turquoise Health, Serif Health) and open efforts (DoltHub — collection stalled since 2023) have already parsed this data at scale. The open, unsolved problem is maintenance: a continuously refreshed, openly redistributable national dataset. DoltHub has publicly offered to host exactly that if someone collects it.
⚡

Decision sequence

  • 01Before any scheduled procedure: find your hospital's cms-hpt.txt at hospitalwebsite.com/cms-hpt.txt — it links the official price file. Or use the open-source discover.py to do it in one command.
  • 02Look up your CPT code (it's on your order/referral): query sticker, cash, and your insurer's negotiated rate, in the right setting (inpatient vs outpatient).
  • 03If the cash price is below your plan's rate and you haven't met your deductible, ask for the cash price in writing — it is published, and binding as the self-pay rate.
  • 04Benefits teams: pull the 300 codes that drive 80% of your spend, load them for your contracted hospitals, and compare. The whole national dataset fits on one machine.
  • 05Contribute: run the parser on your local hospital and publish the stats — the toolkit is MIT-licensed and designed to be run by anyone.
⚠️

Cost of skipping this

  • ✕Sample bias: 5 hospitals in 4 states, chosen for file accessibility, not representativeness. Spreads this large are unlikely to be artifacts, but the national distribution is unmeasured here.
  • ✕Self-reported data: hospitals publish what they publish; CMS enforcement of accuracy is thin. Files can lag real contract changes.
  • ✕Percentage/algorithm contracts are excluded by law from dollar disclosure — true negotiated prices are partially unobservable in the files.
  • ✕Case-rate artifacts (corrected in Rev 1.1): identical dollar case rates span hundreds of unrelated CPT codes and inflate naive maxima — a first-pass 101.8x claim here traced to this. Spread statistics now exclude amounts shared by 50+ distinct codes; codes contracted only as case rates are labeled as such.
  • ✕Prices are episode rates for the hospital component; physician fees, anesthesia, and implants may bill separately. Out-of-pocket depends on plan design.
  • ✕Not medical advice, and not a substitute for a Good Faith Estimate — but it is a cross-check the industry hopes you won't do.
✋

Who this report is NOT for

  • ✕Hospitals and payer contracting teams — you already have this data contractually; this is a patient-side tool
  • ✕Anyone needing a guaranteed out-of-pocket estimate — plan design, deductibles, and physician/separate-biller fees mean this data cannot predict your exact bill

Honest disqualification. If none of the above matches you, this report was written for you.

What's Inside

4 deliverables
database

10.1M-row parsed dataset

Five hospitals' complete price files, normalized and queryable

code

Open-source toolkit

discover / ingest / query / stats — MIT, runs on a laptop

check

Verified numbers

Every figure reproducible via report_stats.py

book

Patient playbook

How to pull your hospital's real rates before a scheduled procedure

Full Report

All 4 sections — scroll down to read.

4 sections
01The law works. The format defeats it.free
0277x sticker-to-negotiated, within one hospitalfree
03Same knee, 7.7x price differencefree
04The entire country fits on your laptopfree
01

The law works. The format defeats it.

The CMS Hospital Price Transparency rule, in force since 2021 and strengthened for 2026, is one of the most complete disclosure regimes in American healthcare: every hospital must publish, in a machine-readable file, its gross charge, discounted cash price, and every payer-specific negotiated rate for every service. Compliance, after early resistance, is now broad. And it almost doesn't matter, because the median file is hundreds of megabytes of JSON — one we measured is 843 MB, which is larger than the entire text of Wikipedia when it was published. No spreadsheet opens it. No browser renders it. The law delivered the data and the delivery format destroyed the information.

02

77x sticker-to-negotiated, within one hospital

Take a brain MRI with contrast (CPT 70553). At Cedars-Sinai, the sticker price is $15,417 while the per-code negotiated rates insurers actually pay, in the same outpatient setting, run from $445 to $8,675 — a 19.5x spread on the same code, same setting, same building, before any cross-hospital comparison. CT abdomen with contrast at the same hospital: $17,058 sticker against a best negotiated rate of $222 — 77x. A wrist X-ray (CPT 73630) negotiates at $17.30 at UCLA and $469.94 at Providence Washington — 27x across town. Correction in this revision: our first pass reported spreads up to 101.8x, but those maxima traced to flat case rates published identically across hundreds of unrelated codes; Rev 1.1 excludes them, and every number in this section survives that filter. The chargemaster figure that dominates billing disputes is still routinely an order of magnitude above what the system actually transacts at.

2 more sections in this report

What unlocks with purchase:

  • 03

    Same knee, 7.7x price difference

  • 04

    The entire country fits on your laptop

One-time purchase · Instant access · No subscription

03

Same knee, 7.7x price difference

Total knee replacement, CPT 27447, outpatient, negotiated rates: $6,121 at Hoag Memorial Hospital Presbyterian in Newport Beach, CA. $47,391 at Ronald Reagan UCLA Medical Center in Los Angeles — 15 miles away. Both are real contracted rates from the hospitals' own files, published as flat case rates covering the full surgical episode — the form in which hospitals typically contract major surgery — rather than per-line negotiated prices. A patient with a high-deductible plan at the second hospital could owe more than six times the entire cash-package price of the first. Cross-hospital variation of this magnitude is the norm in our sample, not the exception: brain MRI negotiated rates span $311 (UCLA) to $4,402 (Providence Washington) across the five institutions.

04

The entire country fits on your laptop

The five files we parsed total 1.5 GB and 10.1 million rows. There are roughly 6,000 hospitals in the United States. Extrapolating the observed file sizes, the complete national price dataset is on the order of a few hundred gigabytes raw — and compresses to a queryable database well within a single consumer machine. This is the central finding of this report, beyond any specific spread: the engineering excuse for not knowing American hospital prices is over. Any journalist, benefits team, researcher — or patient — can hold the whole market in one database. The parser that proves it is open source, handles both required formats, and ingested the largest file in 34 seconds.

Evidence & Citations

Every claim in this report traces to a verifiable source.

Last reviewed September 21, 2026

CMS Hospital Price Transparency (45 CFR Part 180)
https://www.cms.gov/priorities/key-initiatives/hospital-price-transparency
Accessed September 21, 2026
CMSgov/hospital-price-transparency — official schema & templates
https://github.com/CMSgov/hospital-price-transparency
Accessed September 21, 2026
fairprice — open-source parser & findings (this report's toolkit)
https://github.com/mfethe1/fairprice
Accessed September 21, 2026
eCFR 45 CFR Part 180 — full regulation text
https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-E/part-180
Accessed September 21, 2026
Turquoise Health — free consumer hospital price lookup (prior art)
https://turquoise.health/patients
Accessed September 21, 2026
DoltHub hospital-price-transparency — open dataset, collection stalled 2023
https://www.dolthub.com/repositories/dolthub/hospital-price-transparency
Accessed September 21, 2026
PatientRightsAdvocate.org — Hospital Price Files Finder, 6,000 hospitals
https://www.hospitalpricingfiles.org/
Accessed September 21, 2026

Methodology

Who wrote this, what evidence shaped it, and how the recommendations are framed.

  • ●Parsed hospitals' own machine-readable price files (CMS 45 CFR Part 180 MRFs) — both JSON v2.x and tall-CSV formats — with a streaming parser that handles 843 MB files on a laptop.
  • ●Every comparison is within-setting (inpatient vs outpatient never mixed) and within-hospital unless explicitly labeled cross-hospital.
  • ●Case-rate filtering (new in Rev 1.1): hospitals publish identical dollar case rates across hundreds of unrelated CPT codes. Amounts shared by 50+ distinct codes are excluded from spread statistics; our first-pass 101.8x maximum traced to exactly this artifact.
  • ●All findings reproducible: the parser, query tool, and stats generator are open source at github.com/mfethe1/fairprice.
  • ●Sample: 5 hospitals across 4 states (CA, MA, WA), 10.1M price rows, files self-reported 2025-2026. Explicitly not nationally representative.

Author: Rare Agent Work · Written and maintained by the Rare Agent Work research team. All prices sourced from hospitals' own CMS-mandated filings.

Why This Report Earns Attention

Proof 1

10.1 million price rows parsed from 5 hospitals' own CMS filings, including both required file formats.

Proof 2

Within-hospital spread: at one hospital a brain MRI with a $15,417 sticker price carries per-code negotiated rates from $445 to $8,675 — a 19.5x gap. Across the five hospitals the span runs $311 to $8,675.

Proof 3

Cross-hospital spread: the same knee replacement (CPT 27447, outpatient) is contracted at $6,121 at one hospital and $47,391 at another — 7.7x. Both are flat case rates covering the full surgical episode, not per-line negotiated prices.

Proof 4

Cash-price honesty varies wildly: one hospital's cash price is 3% of sticker; another's is 100% of sticker.

Ask the Implementation Guide

Powered by Claude — trained on this report's content. Your first question is free.

Ask anything about implementation, setup, or how to apply the concepts in this report. Your first question is free — then we'll ask you to sign in.

Powered by Claude · First question free

When the report isn't enough

Bring a real problem for direct human review.

Architecture review, implementation rescue, and strategy calls for teams with real blockers. Every intake is read by a human before any next step.

Start an AssessmentBook a Strategy Call

Also from Rare Agent Work

Free · open access

Agent Setup in 60 Minutes

Low-code operator playbook for first-time builders

Free · open access

From Single Agent to Multi-Agent

How to scale from one assistant to an orchestrated team

→

Need help deploying?

Book a free workflow audit

© 2026 Rare Agent Work · Home · Reports · Methodology

livenew:LLM-based classifier is 96% accurate but fails on the 4% that matters most194d ago · post yours · rss