How long a wrong listing stays wrong

A count says how many listings contradict the federal record today. It does not say whether anything is being fixed. Age does.

By the Rollcall Health research team · 21 August 2026

Every finding we raise is date arithmetic. A federal file recorded something on a day we can name, and a published directory still contradicted it on the day we read it. Subtract one from the other and the finding has an age. Across the 104 Individual-market medical directories on the CMS federal Marketplace, read on 30 August 2026, we can date 14,803 of them. The median had been true for 327 days.

The spread matters more than the median

Of the findings we can date, 52% were under a year old, 38% between one and five years, and 10% had stood for more than five years. The oldest had been true for 19.9 years. Those percentages describe the backlog by age. They say nothing about how large the backlog is against the size of a directory, which is a separate measure and lives on the Index.

The under-a-year group is roughly what a working process looks like. Records change, a refresh catches them, and some are in flight on the day we read. The group above five years is harder to explain that way, because no refresh cycle is slow enough to produce it. A tail that long is hard to reconcile with a refresh that works consistently, and suggests some records are escaping whatever control is in place. What we cannot say is why, because we are reading the output and not the plan's workflow.

Exclusions run older than deactivations

Split the same findings by the federal file behind them and the two halves come apart. A deactivated NPI still listed had a median age of 327 days. A provider on the HHS-OIG exclusion list still listed had a median age of 740 days, across 340 dated pairs.

One explanation fits the shape of it. The deactivation report is a bulk file a directory pipeline can diff on a schedule. The exclusion list is a compliance instrument, and exclusion screening tends to concentrate around credentialing, on a roster, against the people a plan believes it has under contract. A published directory holds a different population from a roster, so a check that runs against one need not reach the other. That is inference from the age gap, not something we can observe inside a plan.

What an age does not prove

Age is not a violation. A finding says that a published listing contradicts a dated federal record, and how long that has been the case. It does not establish that anyone broke a rule, that any member was harmed, or that a claim was paid, and we cannot see claims at all.

It is also not evidence about a network. Everything here concerns providers whose federal registration was cancelled or who appear on a federal exclusion list. A real, licensed, unexcluded clinician who does not answer the phone, or who will not take the plan, is invisible to both of our checks by construction. Settling that takes contact rather than a file join, and we keep the two apart on the method page.

One limit is worth stating plainly, because it is the first thing anyone asks. We hold the age of a finding. We do not hold a copy of what a directory said on some arbitrary past date. Each finding's lifecycle is dated and reproducible from the federal files; a snapshot archive of published directories is a different asset, and we do not have one.

Why duration is the number to watch

A count is a photograph. Two plans with identical counts can be in completely different condition, one clearing findings in weeks and the other carrying the same rows for years, and a count will not separate them. Duration will, which is why our monitoring reports it per finding rather than as a total that resets every week.

It is also the measure that lines up with what plans are asked to do, though the obligations differ by programme and this page reports the Marketplace. A qualified health plan owes an up-to-date, accurate and complete directory under 45 CFR 156.230, which sets no interval of its own. The 30-day update duty people usually quote sits in the Medicare Advantage rule at 42 CFR 422.111(m), and the verification cycles arriving over the next two plan years are counted in days as well. A median in the hundreds of days is not a compliance verdict against any of them. It is the right unit for the argument.