Ninety-Nine
Point Six

Health Tech Nerds reported this week that San Antonio's 28,000-life city employee plan is running roughly $40 million over a $250 million budget, and that a four-location freestanding ER chain called Prestige accounts for about 53% of citywide ED claims while winning its federal arbitrations 99.6% of the time.

That number is checkable. Every determination the No Surprises Act's Independent Dispute Resolution process makes is published, line item by line item, in a CMS public use file. So we checked.

So the reported figure survives contact with the primary data, to the decimal. What the file adds is context the story couldn't: Prestige is not an outlier. Across the same six months, 18,850 provider entities pushed 2,649,878 line items through federal arbitration, and providers won 87.7% of them. Plans offered a median of exactly 1.00× the QPA. Providers asked 4.53×. The winning offer landed at 3.98×.

Everyone who filed at least eight times

Each dot is one provider entity as it appears in the federal file. Horizontal is how many line items it pushed through — a log axis, because the range runs from single digits to three hundred thousand. Vertical is the share the arbitrator decided its way. Dot size is the median winning offer as a multiple of the QPA. Prestige's four locations are ringed.

The stress test

Drag Min line items to the right and watch the top of the chart empty out. That is the whole lesson of this dataset, and it is why a headline win rate is close to meaningless on its own.

In the full file — all 18,850 entities, not the 524 plotted above — 9,440 of them have a perfect 100% record. Here is where they live:

Which means the honest reading of “99.6%” is not this chain is extraordinarily good at arbitration. It's this chain filed enough times for its win rate to be a real number, and the real number is roughly what a well-run repeat filer gets. Prestige's 99.1% across 2,566 line items sits above the 87.7% national average, but so do dozens of entities in the chart above, most of which nobody has written about.

What this can't tell you

• Multiples are not dollars. The file publishes offers as a ratio to the QPA, not as amounts. A 3.0× award on a low QPA can be less money than a 1.5× award on a high one. The $886 → $1,846 per-claim figure in the Health Tech Nerds story comes from the San Antonio plan's own accounting, not from this file, and we could not verify it here.
• Entity names are free text. Providers type their own name into the portal, so “East Prestige Emergency Room” and “EAST PRESTIGE EMERENCY ROOM” are two rows — misspelling included. The same is true across the file, which means the true concentration is higher than any name-based count, including this one. We treated each spelling as its own entity rather than guessing at merges.
• 524 dots, 18,850 entities. The scatter shows the 240 highest-volume entities plus a random 279 drawn from the 8–149 range, plus Prestige's four locations. It is a sample built to make the sample-size effect visible, not a census. The bar chart underneath uses the full population.
• One vintage, two quarters. Jan–Jun 2025. Prestige reportedly began filing aggressively in August 2025, so the surge the story describes is after this window — what you see here is the baseline it was already running.
• Line items, not disputes. A batched dispute can carry hundreds of line items with the same outcome, which inflates volume for anyone who batches well. We excluded component line items and count at the DLI level; disputes-level counts run about 2.5× smaller.
• Ratio columns are dirty. CMS's percent-of-QPA fields contain values up to 2.5 million and a few negatives — almost certainly entry errors. Medians here are computed on the 0–200× band and roughly 18% of rows have no QPA ratio at all.