The Centers for Medicare & Medicaid Services (CMS) selected 590 Medicare Advantage contracts for payment year (PY) 2024 contract-specific Risk Adjustment Data Validation (RADV) audits.
CMS scheduled the Enrollee Data List (EDL) to post on September 11, 2026, in Section 3 of the PY2024 Methods and Instructions. The medical record submission window opens September 18 and closes February 5, 2027. Hardship exception requests are due February 19, 2027. The dates are in Section 3 of the Audit Methods and Instructions.
The EDL Sampled Enrollee List gives you the sampled enrollees, the audited Hierarchical Condition Categories (HCCs) and the diagnosis codes behind them. The EDL Sampling Frame gives you a RADV enrollee ID for every enrollee in the frame and a flag for the ones sampled. Section 5.3.
It doesn't tell you which records to send. It doesn't tell you whether your sample came in rich or lean against your frame. And it doesn't tell you what either one is worth.
The sample is a simple random sample within your contract. Section 4.3. Section 9.2 takes the average change in risk score across the sample and multiplies it by the sum of county rates for every enrollee in the frame. So each sampled enrollee enters that average at roughly one over n.
Your sample size is on the EDL. At 35, each enrollee's weight is 2.86 percent. At 200, half a percent. In dollars, every dollar of error found on one sampled enrollee is scaled up across the frame by roughly N over n, your frame count divided by your sample size. Both numbers are on the EDL.
For PY2024, the confidence interval doesn't reduce the amount. Step 4 decides whether CMS extrapolates. Step 5 sets the amount from the point estimate. A wider interval only matters if it's wide enough to reach zero.
Record selection. Section 6 requires at least one medical record for each audited HCC. It lets you submit up to twice the number of audited HCCs for each sampled enrollee, in any combination. Under 42 CFR 422.311(c)(2)(iv) you can appeal one record per HCC. So the budget is a choice about where to spend records, and the appeal is a choice about which one stands. I rank candidate records on your claims and pharmacy data. Pharmacy data can point you to a record. It can't stand in for one.
Sample against frame. Can the frame IDs be tied back to your own members? If they can, I compare your sample to your frame directly. If they can't, the frame side is an estimate, and I'll tell you how wide it is.
Two dollar figures. Section 9.1 prices the sample. Section 9.2 prices the frame. Section 9.3 says CMS will collect the sample figure when results issue and reserves the frame figure for later, if legally permissible. Section 2 says CMS has not yet decided which it will collect. The 2023 rule behind extrapolation was vacated by a federal district court in 2025, and that decision is on appeal (Section 2, footnote 6). They're different numbers with different uses, and I build both. The frame figure stays a range until abstraction results come back.
Reruns. Footnote 25 says CMS will refresh the sampling frame before calculating results. The frame you read now may not be the one the payment error is computed against. The work reruns when that happens, and again as results land.
I don't code medical records and I don't review charts. Your coders do the chart review, and the work tells them which records to review first. If you need more coders, I can introduce you to a coding partner. I'm not a licensed financial or legal advisor, and nothing here is legal or financial advice.
Whole book, scoped to your payment year 2024 audited contracts and fixed at signature. The build is pooled across contracts because it is the same build. The output is separate for each contract, because the Section 9.2 gate, the sum of county rates and the extrapolation decision are all per contract.
The work runs on your claims and pharmacy data, so it starts once a Business Associate Agreement is in place. The record decisions come late in the submission window, so there's time. BAA review is the part with its own clock.
Criterion 1 in Section 4.1 requires continuous enrollment in the audited contract from January 2023 through January 2024. That puts a ceiling on how much of a contract's membership can reach the frame. I calculated it for all 590 audited contracts from public data. CMS publishes the disenrollment input for 474 of them.
Once your frame is on the EDL, the calculator is a check on it rather than an estimate of it. Two things to know. The ceiling uses a national mortality rate, so if your members die at a materially lower rate, your true retention can sit above it. And a member moved to a different contract number by a consolidation fails criterion 1 without leaving or dying.
CMS's Audit Schedule, published August 28, 2026, puts PY2023 initiation in November 2026, PY2022 in January to February 2027 and PY2025 in April to May 2027. CMS notes the dates are subject to change.
Criterion 1 is the same in substance in every published year from 2018 to 2024. So the ceiling can be computed for a payment year before its audit opens.
Not much else carries over. PY2018 and PY2019 ranked enrollees in the top decile. PY2020 on uses the top quartile. The special needs plan exclusion appears in PY2021 and PY2024. PY2018 ran two frame definitions at once, and only one of them used the ranking. PY2018 and PY2019 sampled 35 enrollees per contract, and the stratified sizes start in PY2020. And PY2018 extrapolated at the lower bound of the confidence interval, where later years use the point estimate. The 2023 rule commits CMS to no single method, only to one that is statistically valid and well suited to the audit (88 FR 6643).