However, risk-limiting audits results require more than the vote tallies as evidence. They involve manually inspecting a random sample of records from a voter-verifiable paper trail of voters’ preferences (ideally, voter-marked paper ballots), and require evidence that the paper trail is complete and intact (Stark and Wagner, 2012). To conduct a risk-limiting audit, the auditors must be able to draw a random sample of ballots and must have physical access to the selected ballots. The probability calculations involve worst-case assumptions about election results--they provide guarantees whether errors are random or are introduced by a malicious opponent--and are based the known (by fiat) probability that the audit will select any particular set of paper records for inspection.