# Risk-Limiting Audit A **risk-limiting audit (RLA)** is a post-election check that compares a random sample of paper ballots, counted by hand, against the machine-reported result. It keeps drawing ballots until the evidence reaches a preset confidence that the reported winner really won. If it never does, the audit escalates to a **full hand count**. The name comes from the guarantee. The **risk limit**, often 5% or 10%, is the maximum chance that a wrong outcome survives the audit uncorrected. The sample size depends on the margin: a landslide can be confirmed with a few hundred ballots, while a razor-thin race can escalate to a full count. The method was formalized by UC Berkeley statistician **Philip Stark** at the end of the 2000s. ## Three methods The [[Election Assistance Commission]] describes three variants: - **Ballot-polling:** tally a random sample of paper ballots and test whether the reported winner leads in it. - **Ballot-comparison:** compare individual paper ballots with the voting system's record of each ballot. This is the most efficient, but it needs equipment that keeps per-ballot records. - **Batch-comparison:** compare hand counts of whole batches with the system's batch subtotals. All three depend on a **[[Voter-Verified Paper Audit Trail]]**. Without paper the voter has checked, there is nothing independent of the machine to audit. ## Where they are used - **Colorado** ran the **first statewide RLA** after the November 2017 election. - **Georgia's** 2020 presidential audit escalated to a **full hand tally of about 5 million ballots**. It confirmed Biden's win; his margin moved to 12,284 votes after uncounted memory cards turned up. It is the best-known RLA in American history, and the one election deniers least accept. - **Today** seven to nine states require RLAs, depending on how strictly "require" is defined. As of mid-2026, the Movement Advancement Project counts Colorado, Georgia, Maryland, Nevada, Rhode Island, Texas and Virginia, with pilots in Connecticut, Indiana, Michigan and New Jersey. Verified Voting's 2025 list adds Pennsylvania and Washington. Maryland's legislature passed its RLA requirement in April 2025. ## Why it became a target RLAs are the strongest available technical answer to claims that voting machines were tampered with. They don't trust the machine; they check it against paper using statistics. That is why the [[Election Denialism]] movement has generally pushed **full hand counts** of every race instead. Hand counts are slower and less accurate, and they move the count back to partisan observers. A movement that rejects sampling also rejects the audit's answer. Trump said on **Aug 18, 2025** that he would sign an order ending mail ballots and voting machines. No signed order on voting machines has been found. Abolishing machine tabulation would make ballot-comparison RLAs impossible. See [[How a Ballot Is Counted]] for where the audit sits in the process. ## See Also - [[How a Ballot Is Counted]]: the count an audit checks, and why a paper record has to exist for this procedure to mean anything - [[Post-Election Audit Crosswalk]]: the fifty-state picture this method sits inside — the four audit designs, the timing-against-certification column, and why the count of RLA-requiring states is seven, eight or nine depending on who is counting - [[How a Midterm Election Is Certified]] · [[Election Certification]] · [[Election Assistance Commission]] · [[Verified Voting]] - [[Voter-Verified Paper Audit Trail]] · [[Election Denialism]] · [[Brad Raffensperger]]