2026 Marketplace directory integrity report

How much of what health plans publish still matches the federal record, measured the same way for every directory in the market.

By the Rollcall Health research team · 16 August 2026

This page reports the 2026 plan year of the CMS federal Marketplace, and it is regenerated from the most recent completed run rather than frozen at publication. The run it currently reports was read on 30 August 2026: 104 directories, 1,329 files, and 16,768,952 listings compared against two federal files. A further 4 directories could not be read at all: 3 carrying forward the measurement from an earlier week, dated as such, and 1 with no score rather than a zero.

What was read

Scope: individual market medical qualified health plans certified for the CMS federal Marketplace. The two federal files are named so anyone can check the same artefacts: the NPPES deactivated NPI report (NPPES_Deactivated_NPI_Report_081026_V2.zip) and the HHS-OIG List of Excluded Individuals and Entities published on 10 August 2026. A later version of either file will give a different answer, which is why this page states which one it read.

What was found

Findings sit at three levels. Level 2 is the one that names a listing.

Level 2 counts distinct pairs of directory and NPI. A clinician listed by four issuers is four findings, because four directories each published the row, but a single directory listing the same NPI twice is counted once.

How old the findings are

A finding is dated from the federal act, not from the day we read it. The median level 2 finding had been published for 327 days. Deactivations ran to a median of 327 days and exclusion matches to 740. The oldest had stood for 19.9 years.

Every one of these rows sat through the plan's own loads, credentialing cycles and internal audits after it stopped being true. The count is how many went unnoticed, and the duration is how long each had to be caught in.

How the directories compare

The Index states findings per 10,000 evaluated records, where higher is worse and zero is the floor. Across the market it reads 15.58, and the median directory reads 4.01. The market figure sits well above its own median because the worst directories carry enough findings to pull it there.

The spread between directories is what this report is really reporting. The cleanest reads 0.00 and the worst reads 272.22, against the same two federal files, read on the same day, under the same rule. Since the measurement was identical, what differs is how the directories are run. The Index publishes how those rates are spread across the market.

What this does not measure

This is a floor, not an accuracy rate. Two checks ran, both of them joins against a federal act. Phone numbers, addresses, panel status and whether a practice takes the plan are in no registry and were not checked. A directory with no findings here is a directory that passed two checks, not a directory that is correct. The method page sets out how each check works.

No plan is named on this page. Each plan's own rows go to that plan first.