Mockingbirds

Every network sings the same song about its discounts. Every hospital sings the same song about what it means to its community. Both songs are usually sincere, neither has been checkable, and the employer paying for all of it has been choosing between them on faith.

My local hospital sold.

Fauquier Health sits in Warrenton, forty-five minutes from Washington, and in 2026 it agreed to join Bon Secours Mercy Health [1]. I do this for a living, and the curious version of me wanted a snapshot: what does this hospital charge, and what does it actually get paid, on the day before it becomes part of somebody else’s system.

Before I get to what I found, I want to be careful about where I point. Both of those claims are usually sincere, both have been effectively unfalsifiable, and neither party had much reason to stop making them.

Most plan sponsors already know that hospital prices vary widely. Few have had an easy way to see by how much, in their own geography, for free. Everyone can now, because of the machine-readable file: a spreadsheet every hospital in the country has to publish, listing what it charges and what each insurer has agreed to pay. It is public, it is free, and almost nobody opens it.

What follows is a cross-walk of what five insurers were actually paid at nine Virginia hospitals for the same handful of conditions. On one sepsis admission, a single carrier’s paid amount runs 3.8 times higher at one building than at another. It was put together in an evening, out of those files, starting the day the sale was reported.

Two things we have been told

Call your local hospital today and ask what a service costs. The answer is almost always the same: it depends on your insurance.

For a decade the received wisdom has run on two tracks, and both are supported by real work.

The first is that rural hospitals are failing. Chartis counts 417 rural hospitals vulnerable to closure and 41.2% operating in the red, with 206 closed outright or converted away from inpatient care since 2010 [2].

The second is that acquisition raises what care costs. Lewis and Pflum examined 81 out-of-market acquisitions and found prices at the acquired hospital rose roughly 17%, with nearby competitors up about 8%, and the effect larger when the hospital being bought was a small one [3].

Both are credible. Neither tells you what happened at any particular hospital. Prices were not public, so whether a specific deal helped or hurt a specific community was argued from position rather than from record.

That is the part that changed. It is most of what this piece is about.

Ten years, briefly

Fauquier carried roughly $90 million of debt when it sold 80% of itself to LifePoint Hospitals in a $250 million deal that closed in October 2013. The debt went away and the PATH Foundation was capitalized with more than $100 million [4]. By 2017 the operating margin had gone from 2.87% to 17.9% while full-time staffing fell from 771 to 631 [5]. LifePoint bought the community’s remaining 20% in 2018, and Apollo Global Management took LifePoint itself private later that year [6].

This is not a distressed rural hospital. It is a profitable one, and has been for most of the last decade.

A margin does not move like that on its own. It can only come from price, acuity, volume, or the cost of running the place. Volume is ruled out: staffing fell by 140 and revenue peaked in 2014, lower by 2017 [5]. Unless the county got measurably sicker, what is left is that care got cheaper to deliver, or dearer to buy, or both. That is an inference from an accounting identity, not a finding.

What the record cannot answer at all is what any of it did to the price of care here, because in 2013 there was no price to look at. The obligation to publish did not exist until 2021. The question every employer paying for coverage here would want answered has no baseline. It cannot be asked backwards.

Which is exactly why it is worth asking forwards. The hospital is changing hands again. This time the file exists.

The snapshot, and why anyone can take one

I did not need a data vendor, a relationship, or anybody’s permission to take that snapshot. Since January 2021 every hospital in the country has had to publish its negotiated prices under 45 CFR 180. And since January 2026, under the CY2026 outpatient payment rule, a hospital that prices a contract as a percentage of billed charges has also had to publish the median amount it was actually allowed on that service, with the 10th and 90th percentile and the number of claims behind them, drawn from its own remittance data over a twelve to fifteen month lookback [7].

That second requirement is the one that matters. A negotiated rate is what a contract says. A median allowed amount is what the money did.

So I pulled nine nearby Virginia hospitals rather than one. Fauquier, two Bon Secours facilities, two Inova, two Sentara and two UVA, captured August 20 and 21, 2026. Together they carry 354 group commercial contract rates and 127 published allowed amounts across six services: three inpatient admissions, an appendectomy, an echocardiogram and an hour of infusion. Everything below comes out of those files.

The song

Ask three carriers why their network is the right one and you will get the same answer in three voices. Deeper discounts. Broader access. Better unit cost. The supporting exhibit is a repricing exercise: take twelve to twenty-four months of the employer’s claims, run them through the competing fee schedule, report the difference.

The exercise is arithmetically sound and structurally blind. It holds facility mix constant, so it assumes members keep going where they went. It runs on history, so it assumes no contract in either network renegotiates inside the projection period. And it treats the discount as something the carrier carries with it, the way a buyer carries purchasing power.

The files say otherwise. Median amount actually paid for the identical sepsis admission, same insurer, different building:

Median allowed amount, MS‑DRG 871, as published by each hospital. Group commercial plans only: exchange and marketplace products, federal employee plans, out-of-state Blues, union supplementals and single-employer plans are all excluded, because this is a question about Virginia employer coverage. The carrier is held constant here; the network product is not, because the nine files do not label HMO, PPO and POS consistently enough to match one to one. Where a carrier files more than one group product at a building the figure averages them. Within a single building the same carrier’s HMO and PPO medians differ by a median of 1.02 times across 30 such pairs, so product mix is part of these spreads and not the whole of them. Nine hospital machine-readable files under 45 CFR 180.50, captured August 20–21, 2026.

The carrier that is cheapest in one building is among the dearest in another, for the same admission, in the same state, in the same year. A discount is not a property of the insurer. It is a property of the insurer at a building, and your exposure to it is a property of where your people go.

The obvious objection, and a way to test it

The first thing anyone should say to that table is that it is not measuring price at all. A sepsis admission at a large academic referral center is a sicker patient than a sepsis admission at a small community hospital. Same DRG, longer stay, more outliers. Of course the number is bigger.

That objection is correct in principle and it deserves better than a caveat, because this dataset happens to contain a way to size it.

Inova Fair Oaks and Inova Fairfax file under one tax ID and publish the same 21,427 rows in the same order, sharing one chargemaster outright. Their negotiated rates are close but not identical. Across the services tested here the two files’ dollars differ by a median of 1.2% and at most 15%, with Anthem and UnitedHealthcare carrying facility-specific carve-outs and the rest matching exactly. So a gap between those two buildings in what was actually paid is mostly not price. It is case mix, length of stay and outliers, very nearly isolated.

Across 15 payer-and-service pairs where both Inova files publish an allowed amount on a group commercial plan, that gap runs to a median of 1.01x, a mean of 1.02x and a maximum of 1.09x [8]. At the top end that is smaller than the rate difference itself, which is the sense in which case mix adds very little here.

Why mockingbirds

Mockingbirds have no song of their own. They repeat what they have heard, convincingly, and they are not lying when they do.

The carrier’s claim has no fixed content, because it describes a relationship between two parties evaluated against a third variable that nobody controls and nobody forecasts.

The hospital’s claim about its community has the same problem in a different key: it is true in the ways it is measured and silent about the ways it is not. The 2013 transaction here took the operating margin from 2.87% to 17.9% while staffing fell by 140, and the same institution kept the doors open in a county that would have felt it if they had closed. Both of those are the record. Neither is the whole of it.

So who wins? On the evidence here, the building wins the admission outright. On outpatient work it depends on whether there is somewhere else to go, and that varies by service. Either way the employer loses whichever argument it was not having. But that is only true while the songs are all anyone has to go on.

Choosing a carrier on its network discount is a bit like choosing a car on what it does to your insurance premium. The math is real, the quote is real, and one at-fault accident makes the whole comparison moot.

Where leverage actually sits

None of this makes the network irrelevant, and I want to be precise rather than contrarian about it.

Run the same nine files a different way. For each service, take one observation per hospital-and-insurer pair, and ask whether hospital identity or insurer identity explains more of the variation in price. On the three inpatient admissions the building explains 48% to 55% and the insurer 10% to 16% [8]. That one holds however it is cut. Drop any single carrier, or any single hospital, and the building still explains several times what the insurer does. On the appendectomy it turns over, 37% insurer against 22% building, but that result is not as solid: drop Anthem and it reverses. The echocardiogram has too few contracts on file to call either way, and I am leaving it out rather than publish a number that will not hold. Group commercial plans only, on the same standard as the table above.

That is not a story about who negotiates harder. It is a story about substitutes. Nobody shops a sepsis admission. The ambulance goes where it goes, and a network without the county’s hospital is not a network. Where an alternative exists the fee schedule should start to matter more than the building, and the appendectomy is the only service here that shows it clearly. On the strength of one service, and one that reverses when a carrier is dropped, I would treat that half as a direction worth testing on your own claims rather than a finding.

Leverage sits wherever the alternative is worst, and it is not distributed evenly. Some employers genuinely do have bargaining power and most do not. A 15,000-life employer concentrated in one metro is a different counterparty from a 200-life employer spread across four states, and pretending otherwise does nobody any favors. The questions below are worth asking in roughly that order, but which ones you can act on scales with your size.

What to run on your own plan

This is testable on any self-funded population without buying anything. The chart, the nine files and a prompt that rebuilds it out of your own facilities sit on the tool page.

• Pull your top ten facilities by paid dollars from your own claims. Not by visit count, by dollars. In most populations that list is short and the top three carry the weight.
• Open each one’s file. Start at /cms-hpt.txt at the root of the hospital’s domain. It names where that hospital’s machine-readable file lives, and it is the fastest route in.
• Read the median allowed amount, not the negotiated rate. Where the median is present, that is what the hospital was paid. Where the count reads “1 through 10,” the figure is thin and should not be quoted.
• Compare that against the discount your renewal deck claims, at the facilities that carry your dollars. A network that buys well at three of your top ten and badly at the other seven is not a better network. It is a better network for somebody else’s referral pattern.

That comparison is the one the repricing exercise cannot make, because it starts from the assumption that the facility mix is fixed. This runs off the hospital’s file rather than your carrier’s because the payer-side Transparency in Coverage files carry contracted rates and nothing else, with no allowed amounts and no volume [9]. The rest of that comparison is in the source manifest.

The file cuts both ways

One more consequence, and it is the one I would watch.

Until this year, what a hospital was actually paid by a given carrier was known to exactly two parties, and each of them knew only their own half. That is no longer true. The file now shows, in the hospital’s own numbers, what every commercial payer at that building settled at.

Which means the next negotiation has a new exhibit, and either side can carry it. A carrier can open the hospital’s file and ask why it is paying more per admission than the carrier one row down. A hospital can open the same file and notice that somebody is paying more than it is getting. Both readings come off the same page.

Nobody has run that play at scale yet, because the allowed-amount columns have only been enforceable since April 1, 2026. But the interesting thing about a disclosure regime is that it does not choose a beneficiary. It just puts the document on the table, and the party who reads it more carefully picks it up.

Four better questions

Not one of these is about the discount.

1. Where do your dollars actually go? Paid dollars by facility, ranked by dollars and not by visit count. This is where the money in a self-funded plan actually goes, run one level down at the building. If the reporting will not produce it, that is an answer about what the reporting is for.
2. Can you move any of them? Steerage only works if something funds it and somebody navigates. An estimator nobody opens is not a program, and what the funding is worth against your own claims is its own arithmetic. Then the harder version of the same question. Your carrier already knows which facility it pays the most for a given condition and which it pays the least, and it holds quality measures at both ends of that range. Ask what part of that reaches a member at the moment they are choosing, in what form, and how you would know if it had.
3. What does your audit clause permit? Scope, sample size, auditor of your choosing, look-back. This one gates most of the others, because you cannot recover what you are not permitted to examine.
4. What did the repricing exercise assume about facility mix and about contracts renegotiating inside the projection period? Both assumptions are usually invisible and both are usually wrong.

There are another six worth asking, and they matter more or less depending on how big you are and what your contract already lets you do. They are in the companion tool, which ranks all ten by how many of your own dollars each one can reach against whether you currently have the right to pull it.

Fringe Theory is independent and unaffiliated. Views expressed are my own and do not represent those of my employer. Nothing here is legal, tax, medical, or investment advice. Figures described as modeled or derived are my own arithmetic from the public sources listed, not disclosed data.