Hospitals & Insurance Both Publish Price Info; We Checked Whether They Match

Since 2021, hospitals have been required to publish a machine-readable file listing what they charge for every service and what each insurer has agreed to pay for it. That rule is Hospital Price Transparency. Since 2022, health insurers have been required to publish the negotiated rates behind every plan they sell, under a separate rule called Transparency in Coverage. The two rules were written for different audiences and they produce documents that look nothing alike, but underneath they describe the same contracts, filed by two parties who have no reason to coordinate what they say.

Do the two sides agree? Others have checked before us. Turquoise Health, a company that sells access to both kinds of files, published a matching analysis in 2023 and treats a rate confirmed on both sides as a product feature. In 2024, researchers at the Hilltop Institute cross-validated 26 Mississippi hospitals against Blue Cross Blue Shield of Mississippi on maternity services: 16.3% of the hospital-and-code combinations appeared in both sources, and where the sources overlapped, 77.4% of the prices matched to the penny. We wanted to run the join ourselves, on the hospitals we already track, and account for every pairing that fails along the way, with the arithmetic shown.

This run is deliberately small, because the point was to find out whether the join is even possible before doing it at scale. One insurer, Blue Cross Blue Shield of Wyoming. One state, which for us means all 30 Wyoming hospitals in our lineup. 74 common procedures, the kind a patient can price ahead of time: imaging, blood work, office visits, and routine surgeries. 65 insurer rate files, captured from the insurer’s machine-readable index on August 1, 2026, set against hospital files we had retrieved earlier in the summer.

The answer arrives in two parts, and the first part is the larger finding. The comparison is almost never possible. Of 175,080 candidate comparisons, 98.6% failed at least one of the checks a fair comparison has to pass: same facility, same kind of bill, an insurance arrangement we can pin down, and one unambiguous rate on each side. Where it was possible, it was exact. At the one hospital where validation fully worked, every one of 312 comparisons matched the hospital’s published dollar to the cent.

The files

A hospital’s machine-readable file is the hospital’s accounting of its prices. Every service it bills, the gross charge attached to that service, the cash price, and a row for each payer contract the hospital holds, all published by the hospital. Most of this site is built out of those files.

An insurer’s in-network file covers the other side of the same contracts, and it is organized on completely different lines. Rates attach to provider groups, and a provider group is a list of NPIs, the National Provider Identifier numbers that identify a clinician or a facility on a claim. A single file can cover many plans at once (the BCBS Wyoming files here do), and a separate index published by the insurer says which plans each file is supposed to cover.

That difference is where the trouble starts. The hospital’s rows name plans. The insurer’s prices attach to provider groups and never to a plan, because the format has no field for one. Neither format carries a key that joins one to the other. That gap is the design of the federal formats, not a choice any one insurer made.

Why 98.6% of comparisons are impossible

The entire sweep sits in one table. A candidate comparison is any pairing where a hospital row and an insurer rate could plausibly describe the same service at the same hospital. Of the 175,080 candidates, 2,425 passed every check, and 828 of those reached a verdict.

We ran it twice on purpose. The strict pass additionally requires the care-setting labels on both sides to match; the relaxed pass treats that label as context, because it largely reflects which rows our extractor kept rather than which service was performed. Everything quoted below comes from the relaxed pass, and the direction of the result is identical in both.

Three blockers account for most of what never gets a verdict.

88,131 pairings put a clinician’s rate against a facility charge. When a hospital posts a price for a CT scan, that price is what the building bills; the radiologist who reads the images bills separately, under a different billing class. Setting one against the other is a category error rather than a price difference, so those pairings are refused.

44,611 involve NPIs that resolve to a hospital while the insurer’s file names a different program that the same hospital runs. The clearest case in this data: a set of NPIs keys straight to Cheyenne Regional’s facility, and the file’s name for that provider group reads “CHEYENNE REGIONAL HEALTH FAIR”. The identifiers say hospital and the label says something else, so we reject the pairing instead of guessing which one to believe.

36,480 fail the plan-scope check, which is the subject of the rest of this section.

Of the 175,080 pairings, the hospital’s side names a specific plan in 107,944 of them and says “all plans” in 36,593, and the remaining 30,543 have no hospital row to compare against at all. Those plan names sound like they ought to help, but they do not, because there is nothing on the insurer’s side to match a plan name against. One insurer file in this run spans 370 reporting plans from 61 sponsors, and no rate inside it is specifically attributed to any one of them.

The receipt for that is blunt. One file publishes both 75% and 95% for a single hospital’s single provider group, on the same code, in the same billing class and the same care setting, and it does so on all 63 procedures we could check there. Two prices, no marker saying which plan gets which. Plan-level attribution is simply unavailable from these documents.

Where it does work, it works exactly

Memorial Hospital of Sweetwater County, in Rock Springs, is where the comparison worked. 312 comparisons across 63 procedures, and every one of them lands on the hospital’s published dollar to the cent.

Here are three of the 312, worked end to end:

Read a row left to right. The hospital publishes a gross charge. The insurer publishes a percentage against that hospital’s provider group. Multiply the two, and the result is the rate the hospital itself published for that insurer, to the penny.

One feature of that arithmetic bounds what the agreement proves. No verdict in this run came from two dollar amounts printed side by side. Every agreement rests on multiplying the hospital’s gross charge by the percentage the insurer states, and we permit that only after corroborating, from the hospital’s file, that its published rates sit at one constant share of its charges. Where that corroboration fails, the row is refused rather than assumed.

Agreement this exact confirms that both parties publish the same contract term. It does not verify what any claim actually paid.

The second hospital that reached verdicts is a different story, and we are not telling it yet. Every one of its decided comparisons showed the two filings stating different terms. We are not naming the hospital or the size of the gap, for two reasons: a hospital’s posted figure can be a placeholder rather than a contracted rate, and the insurer’s file never says what its percentage is a percentage of.

Both files are public, and anyone can redo this arithmetic. What we are withholding is our conclusion, until a paid claim, an explanation of benefits, or the hospital itself corroborates one side.

There is a second thing these files turn out to do, and it is not validation. At five hospitals, every pairing that passed the checks found an insurer-published term with no counterpart anywhere in the hospital’s file. The insurer’s filing supplies what the hospital’s omits.

What the negotiated rate turns out to be

Sweetwater’s entire commercial institutional rate sheet is a single number. 95%, on all 19,539 rows filed against that provider reference. One percentage, applied to whatever the hospital charges.

That is easy to skate past, so it is worth spelling out what it means for a patient. The negotiated discount floats on the chargemaster. If the hospital raises its gross charge, the negotiated price rises right along with it, and the discount stays exactly where it was.

Across BCBS Wyoming’s files, institutional percentage terms run from 65% to 100%, spread over 22 distinct values.

Cheyenne Regional is the counter-example. Flat percentage terms cover 34 of its imaging, endoscopy, therapy and office-visit procedures. Thirteen lab tests carry true per-procedure dollar amounts instead: 80061, a lipid panel, at $20.09; 84443, a thyroid test, at $25.20; 81002, a urinalysis, at $5.22. Those dollars are one of two schedules the insurer files for that hospital; a separate file publishes a systematically lower amount for the same tests. A few contracts with one insurer cannot tell us how negotiated rates work in general. Each one here has its own structure, so we are describing them rather than asserting a single common pattern.

The false 20-point overcharge we didn’t publish

This method has a failure mode, and it nearly took us.

An earlier pass found what read like the hospital’s BCBS contract stating 75% where the commercial contract says 95%. A 20-point gap between two of the same insurer’s filings for the same hospital would have been a story.

The tell was not in the rate file at all. It was in the index. The plan_sponsor_name on that file was the hospital itself, which makes it the hospital’s employee benefit plan, sponsored for its staff and administered by the same insurer. That is a real contract, and it has nothing to do with what the hospital bills the insurer’s commercial members. The same shape turned up at a second hospital.

We built a 20-point disagreement that does not exist, and the only thing that caught it was reading the index instead of trusting the rate file. Naive versions of this comparison will confidently mislead.

What this run does not establish

• This was just one insurer, one state, and 74 common procedures. Nothing here generalizes to another payer or another state.
• The two sides were not captured together. The hospital files were retrieved between June 18 and July 14, 2026, the insurer’s files came from its August 1, 2026 index, and the comparison ran on August 11. A contract term could have changed in the weeks between the two captures, so drift is a bigger caveat here than it would be on a same-day snapshot.
• Percent-of-charges contracts are not universal even in Wyoming. Two hospitals which reached the base check publish fee schedules instead.

Both sides file these documents under federal rules, and the arithmetic on them is ordinary multiplication. The insurer’s index is public and names every rate file we read, though the links it hands out are signed and expire. Our sources and method are laid out on the about page.

Correction, August 20, 2026: this article originally said that, as far as we could tell, nobody had publicly checked whether the two sides agree. Two prior efforts had: Turquoise Health’s 2023 matching analysis and the Hilltop Institute’s 2024 Mississippi cross-validation, both now cited in the introduction.

See something wrong? Tell us →