Nebraska's Medicaid director keeps returning to one number: roughly 74% of enrollees confirmed compliant with the new work requirement without being asked for a single document. That capability — clearing people from data you already hold — is already measured, for renewals, in every state. It runs from 11.8% to 80.6%. Here are all 51, and 69 million renewals.
Nationally, of every 100 renewals that came due: 55 cleared automatically, 22 came back a form, 12 lost coverage because paperwork didn't close, 7 were found genuinely ineligible, and 3 were still pending. The state that automates well isn't denying fewer people. It's asking fewer people.
Each dot is a state. Area is renewal volume. Drag the axis brushes to filter, drag the volume slider to drop small states, and watch the correlation move.
Drag min renewals due to zero and hit show only n<300k. In 2024–25 the ten smallest states produce r = −0.84 — a beautiful, publishable-looking finding that more automation means less paperwork loss. Now switch the period to the unwinding. Same ten-ish states, same metric: r = −0.08. The relationship evaporates.
Across all 51 states the honest number is stable: r = −0.49 post-unwinding and −0.44 during it — real, moderate, and nothing like a mechanism. Ex parte capability explains roughly a quarter of the variance in who loses coverage on paperwork. The other three quarters are notice design, call center capacity, response windows, and how hard the state tries.
This is the federal government's own eligibility-operations dataset. It has 22 columns. It tracks renewals initiated, renewals due, ex parte renewals, form renewals, disenrollments, ineligibility findings, procedural terminations, pending renewals, and a footnote field for each. It contains nothing about exemptions.
The medically frail exemption — the off-ramp for people too sick to meet the requirement — is implemented state by state as a diagnosis code list. Nebraska's layers severity signals on top of the codes: not just a cancer diagnosis, but whether it produced an inpatient stay, whether there's a wheelchair, whether the condition is currently doing something to you. Other states' lists, per Gonshorowski, include wrist splints. Or every diabetes code, flat.
Two states, same statute, same phrase, two completely different populations through the door. That is not a policy disagreement — it's a value-set design decision, made by whoever was in the room, and it is load-bearing for tens of thousands of people.
If you have argued about whether a value set should include the unspecified codes, you already know how this goes wrong. You have just never had the argument where losing it costs someone their insurance. And unlike a quality-measure value set, none of these lists live in VSAC with a steward, a version number, or a change log. The scatter above can tell you how well a state automates. Nothing published can tell you what its exemption list catches.
1. Nebraska's 74% is not the 42.8% on this chart. Gonshorowski's figure is the share of enrollees cleared for work-requirement compliance using income data. The x-axis here is the share of renewals due that closed ex parte over twelve months — 42.8% for Nebraska, or 62.0% if you use completed renewals as the denominator. Same machinery, three denominators, three numbers. Ask which one before you quote one.
2. Renewal outcomes are reported as of roughly three months after the renewal month, and states amend them. This uses each state-month's latest submission. Several states carry footnotes for held terminations or partial reporting; those are not modeled here.
3. High procedural loss is not automatically a failure, and low is not automatically success. A state that pauses terminations posts a low number for reasons that have nothing to do with how well it works. The comparison worth making is a state against itself across the two periods — toggle it and watch Texas, Oklahoma and Colorado move a very long way.