Series 19: Compliance Systems vs. Recognition Systems Article 19A
Two state Medicaid directors receive identical letters from CMS. Both states have expansion populations exceeding 400,000 adults. Both face the December 2026 deadline to implement community engagement requirements under the One Big Beautiful Bill Act. Both must design systems to verify that 18.5 million Americans nationwide, and hundreds of thousands in their states, are meeting 80-hour monthly work requirements.
Director Chen reads the letter and calls her operations team. “How do we confirm that people who are already working get recognized for it?” she asks. Her team begins pulling unemployment insurance wage data, cross-referencing SNAP employment records, and mapping employer concentrations in their expansion population. They discover that 68 percent of their expansion adults already show wages in state databases. Another 12 percent are receiving disability benefits. The team starts building systems to match what they already know against what they need to verify.
Director Hargrove reads the same letter and calls his compliance team. “How do we catch people who aren’t meeting the requirements?” he asks. His team begins designing a reporting portal, establishing monthly submission deadlines, and drafting termination notices for members who fail to report. They build the system around the assumption that members must prove their compliance or face consequences.
Eighteen months later, Director Chen’s state reports a 6 percent coverage loss rate. Director Hargrove’s state reports 23 percent. Both states have the same federal requirements. Both states have comparable expansion populations. The difference is not policy. The difference is paradigm.
This is not a hypothetical. It is the difference between what happened in Arkansas in 2018 and what states like Ohio are designing for 2026. The question every state faces is not whether to have work requirements. Congress answered that question. The question is whether to build systems that see compliance or systems that punish the failure to prove it.
The Compliance Paradigm#
The compliance paradigm begins with a default assumption: beneficiaries are non-compliant until they prove otherwise. This assumption shapes every downstream design decision, from system architecture to staffing models to success metrics.
Under this paradigm, the burden of proof falls entirely on the individual. A person receiving Medicaid must affirmatively demonstrate, through documentation submitted to the state, that they are meeting work requirements. The system does not look for evidence that the person is working. It waits for the person to provide that evidence. If the person fails to provide it, the system treats them as non-compliant, regardless of whether they are actually working.
Documentation becomes the gatekeeper rather than the confirmation. The critical transaction is not “are you working?” but “can you prove you’re working?” These are profoundly different questions. The first is about behavior. The second is about administrative capacity. A person bagging groceries for 35 hours a week at two different stores answers the first question affirmatively every week. Whether they can answer the second depends on whether their employers provide documentation, whether they can aggregate hours across jobs, whether they have internet access to submit reports, and whether they understand the reporting requirements in the first place.
Systems designed around the compliance paradigm optimize for enforcement, verification, and termination. Administrative convenience takes priority over beneficiary capacity. The system is built for the administrator’s workflow, not the worker’s reality. Monthly reporting portals, narrow submission windows, and automated termination processes reflect what is easiest for the state to administer, not what is most likely to produce accurate results.
Success under this paradigm is measured by terminations. When people lose coverage, the system is working. Non-compliance has been identified and consequences have been applied. The question of whether terminated individuals were actually non-compliant or simply unable to navigate the verification process does not factor into the success metric. Termination volume becomes a proxy for program integrity.
Arkansas in 2018 represents the paradigm case. The state implemented online-only reporting through a web portal, required monthly submissions within narrow windows, provided minimal outreach about the new requirements, and automatically terminated coverage for anyone who failed to report. The system was designed to catch non-compliance efficiently. It did not consider whether the non-compliance it was catching was real.
The results were devastating precisely because the paradigm was wrong. Research by Benjamin Sommers and colleagues at Harvard, published in the New England Journal of Medicine, found that 95 percent of those who lost coverage were either working or qualified for exemptions. The system was extraordinarily efficient at terminating people. It was extraordinarily poor at determining whether those terminations were justified.
The Recognition Paradigm#
The recognition paradigm begins with a different default assumption: most beneficiaries are already working or legitimately exempt, and the system’s job is to verify that reality rather than to assume it does not exist.
This assumption is not naive optimism. It is grounded in evidence. Research consistently shows that the vast majority of Medicaid expansion adults who are able to work are already working. Kaiser Family Foundation data indicates that among non-disabled, non-elderly Medicaid expansion adults, roughly 60 percent are employed and another 30 percent have legitimate reasons for not working, including caregiving, disability not yet formally documented, school enrollment, or illness. The genuinely non-compliant population, people who could work, are not working, and do not qualify for any exemption, represents a small fraction of the total.
Under the recognition paradigm, the burden of proof falls on the system rather than the individual. The system must demonstrate that it has exhausted available data sources before concluding that someone is non-compliant. Unemployment insurance wage records, state new hire databases, SNAP employment and training records, TANF work participation data, educational enrollment systems, and disability databases all contain information about what expansion adults are doing. A recognition system queries these sources first and only asks individuals to self-report when administrative data cannot confirm compliance.
Documentation becomes confirmation rather than barrier. The system starts with what it already knows and asks individuals to fill gaps, not to rebuild the entire picture from scratch. A worker whose wages appear in unemployment insurance records does not need to separately prove they are working. The system has already recognized their compliance. Only workers whose circumstances fall outside administrative data systems, gig workers, cash economy participants, those with multiple informal jobs, need to provide direct documentation.
Systems designed around the recognition paradigm optimize for data matching, multiple verification channels, and retention. Beneficiary capacity takes priority over administrative convenience. The system is built around the worker’s reality, then adapted for administrative needs. Multiple reporting channels, flexible timelines, and proactive outreach reflect what is most likely to produce accurate classification.
Success under this paradigm is measured by accurate classification. The system succeeds when working people are correctly identified as compliant, exempt people are correctly identified as exempt, and genuinely non-compliant people are correctly identified as non-compliant. Coverage loss among compliant individuals represents system failure, not system success. The relevant metric is not how many people were terminated but how many were correctly classified.
What recognition architecture looks like in practice is visible in Ohio’s proposed approach to work requirements. The state plans to use unemployment insurance wage data to automatically verify employment for an estimated 60 to 70 percent of expansion adults before those individuals submit a single document. Social Security data identifies disability exemptions. SNAP and TANF compliance records confirm participation in other work programs. Only the remaining 30 to 40 percent whose circumstances cannot be confirmed through automated channels enter active verification workflows requiring individual action.
Why the Paradigm Matters More Than the Policy#
The same 80-hour monthly work requirement can produce coverage loss rates ranging from 5 percent to 25 percent depending on which paradigm shapes system design. The policy is identical. The outcomes are radically different. This observation should fundamentally reorient how we think about work requirements.
The Arkansas evidence makes the case starkly. Sommers and colleagues found that 97 percent of those subject to requirements were already compliant through work or exemption eligibility. Yet 25 percent lost coverage. The gap between actual non-compliance (roughly 3 percent) and coverage loss (25 percent) represents system-generated harm. These were not people who refused to work. They were people who could not prove they were working through the specific channels the system demanded.
Compliance systems generate false negatives at scale. In verification terminology, a false negative occurs when someone who is actually compliant is classified as non-compliant. Compliance systems produce false negatives because they treat the absence of proof as proof of absence. If a person does not submit documentation, the system concludes they are not working. But the absence of documentation tells you nothing about the absence of work. It tells you only about the absence of documentation.
The false negative rate in Arkansas was extraordinary. For every person correctly identified as genuinely non-compliant, roughly eight compliant people were incorrectly terminated. In any other verification domain, an 8:1 false negative ratio would be considered catastrophic system failure. In fraud detection, such a ratio would mean flagging eight legitimate transactions for every fraudulent one. In medical testing, it would mean telling eight healthy patients they were sick for every actually sick patient identified. No well-designed verification system operates at this error rate.
Recognition systems minimize false negatives while still identifying genuine non-compliance. By starting with administrative data rather than individual reporting, recognition systems verify compliance for the majority of the population without requiring any individual action. The remaining population that requires active verification is smaller, making it possible to invest more resources per person in accurate determination. The result is lower false negative rates and more precise identification of actual non-compliance.
The philosophical debate about whether work requirements are good policy is important. But it is secondary to the paradigm choice. A person who believes work requirements promote personal responsibility should still prefer recognition systems, because compliance systems terminate working people alongside non-working people, undermining the policy’s stated purpose. A person who opposes work requirements on principle should still prefer recognition systems if requirements are going to exist, because recognition systems minimize harm to vulnerable populations. The paradigm question cuts across the policy debate.
The Political Economy of Paradigm Choice#
If recognition systems produce better outcomes by every measurable standard, why do compliance paradigms dominate? The answer lies in the political economy of verification.
The framing matters enormously. Compliance systems are sold as “fraud prevention” and “program integrity” measures. Recognition systems lack an equally compelling political narrative. Telling voters “we built a system that catches cheaters” is more politically potent than telling them “we built a system that accurately classifies people.” The former implies action against wrongdoing. The latter sounds like bureaucratic process improvement.
There is a deep asymmetry in visibility between the two types of errors these systems produce. False negatives, compliant people incorrectly terminated, are invisible in the political landscape. A worker who loses Medicaid because they could not navigate the verification portal does not appear on anyone’s political radar. They become uninsured quietly. They delay care quietly. They accumulate medical debt quietly. Their story does not make the news, does not generate constituent complaints to legislators, and does not create political consequences for the officials who designed the system.
False positives, people receiving benefits they should not receive, are politically explosive. A single case of someone gaming the system generates more political attention than ten thousand cases of working people losing coverage for administrative reasons. Media coverage, legislative hearings, and political campaigns amplify fraud stories. No equivalent amplification mechanism exists for administrative harm stories.
This asymmetry creates political incentives that favor compliance theater over recognition accuracy. A compliance system that terminates 25,000 people looks tough on fraud even if 24,000 of those terminations were wrong. A recognition system that retains 24,000 working people in coverage risks being characterized as “soft” on enforcement even though it produced far more accurate results.
Shifting this conversation requires reframing what program integrity actually means. Program integrity is not maximizing terminations. Program integrity is ensuring that eligible people receive benefits and ineligible people do not. A system that terminates eligible people is not demonstrating program integrity. It is demonstrating the opposite. It is using taxpayer resources to generate administrative harm, create downstream costs through emergency care and re-enrollment processing, and undermine the program’s purpose.
The question is whether policymakers and advocates can make this reframe politically viable before December 2026. The evidence from Arkansas is unambiguous. The design principles are well understood. The question is whether political incentives will permit states to build what the evidence demands.
The Choice Ahead#
The policy question has been answered by Congress. The One Big Beautiful Bill Act requires community engagement for Medicaid expansion adults beginning December 2026. Whether one agrees or disagrees with that decision, it is the law.
The paradigm question remains open. Every state must decide whether to build compliance systems or recognition systems. Every MCO must decide whether to invest in member navigation or process termination notices. Every CMS official must decide what guidance to issue and what performance standards to enforce.
States choosing recognition will retain coverage for working people, produce accurate classification of their expansion populations, minimize downstream costs from wrongful terminations, and achieve the policy’s stated goals of promoting work while maintaining coverage.
States choosing compliance will replicate Arkansas. They will terminate working people alongside non-working people. They will generate false negatives at scale. They will produce coverage losses that far exceed actual non-compliance rates. They will spend more on re-enrollment processing, appeals, and emergency care than they would have spent on recognition infrastructure.
The evidence is clear. The design principles are known. The only question is whether the people making these decisions will follow the evidence or follow the politics.