Ten quarters of federal out-of-network arbitration. Volume, win rate, and the size of the award have all moved in the same direction, every single period. Nothing in the statute makes them come back down.
Primary source: Kevin O’Leary, “The Impacts of the No Surprises Act: a San Antonio employer goes over budget,” Health Tech Nerds, Aug 21 2026
Data: CMS Federal IDR Public Use File — OON emergency & non-emergency dispute line items, all nine vintages (2023 Q1 – 2025 H1), via mimilabs
The No Surprises Act was supposed to take the patient out of the middle of an out-of-network bill. It did. What it put in the middle instead is a federal arbitration process, and the story out of San Antonio this week — a city plan $40 million over a $250 million budget because one four-location ER chain figured out how to use it — is not a local anomaly. It is what the whole system has been doing since the day it opened.
Here is every reporting period CMS has published, all three numbers at once.
line items decided (per quarter)provider win ratemedian award, × QPA
Q1 2023
62,156
line items decided
Q1–Q2 2025
2,649,878
21× the opening quarter
Win rate
69.8 → 87.7%
decided for the provider
Median award
2.69 → 3.98×
the QPA benchmark
Where the awards actually land
Three quarters of a decade of policy argument has been about one question: does arbitration anchor on the benchmark, or does it drift above it? The file answers directly. Every winning offer is published as a ratio to the QPA — the plan's own median contracted rate for that service in that market. Here is all 2,241,655 of them from the most recent period.
Two populations in one chart. The spike at 1.0–1.5× is 255,104 line items — almost entirely determinations the plan won, because the plan's offer is the QPA. Everything to the right of it is a provider win. 14.5% of all awards land above 12× the benchmark.
Median plan offer · 1.00× QPA — the statutory anchor, exactly
Median provider offer · 4.53× QPA
Median winning offer · 3.98× QPA
Share decided for the provider · 87.7%CMS Federal IDR PUF, OON emergency & non-emergency, reporting period Jan 1 – Jun 30 2025. 2,649,878 dispute line items, component line items excluded. Ratio columns computed on the 0–200× band; roughly 18% of rows carry no QPA ratio and are excluded from the medians, not zero-filled.
The 80/20 lens
Baseball-style arbitration was chosen deliberately: each side submits one number, the arbitrator picks one, and neither can split the difference. The theory was that this would push both sides toward the middle. It hasn't, because the two sides aren't playing the same game. The plan's number is fixed by what it already pays in-network. The provider's number is whatever it wants to charge. One side can move; the other structurally can't.
So the 87.7% isn't evidence that plans argue badly. It's the arithmetic of a contest in which one player's only legal move is the benchmark and the referee is instructed to consider — but not defer to — that benchmark.
And the volume is the actual mechanism. At a mean $635 IDRE fee and 90 business days per determination, 2.65 million line items in six months is roughly $1.7 billion of arbitration administration spent deciding how to price care that has already been delivered. That is the cost before anyone is repriced — and it lands on self-insured employers like a city of 28,000, not on the insurer whose name is on the card.
What this can't tell you
The last bar is two quarters averaged. CMS's most recent vintage covers Jan–Jun 2025 as a single reporting period. We divide by two for the per-quarter chart and label it as an average; the raw six-month total is 2,649,878 line items. If filing accelerated within that window — and every prior quarter says it did — the true Q2 figure is higher than the bar shown.
Line items, not disputes, and not dollars. A single batched dispute can carry hundreds of line items. Disputes-level counts are roughly 2.5× smaller (1,058,514 in the latest period). And a multiple of the QPA is not a dollar amount: 4× a low benchmark can be less money than 1.5× a high one.
Two vintages have no ratio data at all. The percent-of-QPA columns were not published for the 2024 Q3 and 2024 Q4 files, so the navy line is interpolated across that gap rather than measured. It is drawn dashed there for that reason.
Determinations, not filings. The file records disputes that reached a payment determination. Disputes closed for ineligibility, withdrawn, or settled before determination don't appear — and eligibility rejections have been a large share of the process's total intake.
“Winning offer” is not the same as “what got paid.” Post-determination payment compliance is separately reported and imperfect.
The ratio columns contain obvious errors. Values run as high as 2.5 million and as low as −18. Everything here is a median on a bounded band, never a mean, for exactly that reason.
A benchmark the referee may consider but need not follow is not a benchmark. It is a floor.What ten quarters of the public use file describe
⚠︎ AI-generated · not reviewed by a human · verify against the linked sources before relying on it.