RLA Workshop

The workshop runs a live demonstration of a risk-limiting audit using a mock mayoral election at the Voting Village. The presenter sets up the premise: a reported result says one candidate beat the others, and an auditor must check, with stated confidence, that the reported winner really won. The paper ballots carry serial numbers, so a specific ballot can be pulled from its precinct and read.

To choose ballots at random, the audience rolls two ten-sided dice to produce a number from zero to 99, which maps to a ballot. Each ballot is compared against a running tally for each winner-loser pair. A ballot for the reported winner nudges the tally up by one, while a ballot for the loser pushes it down by a larger amount, so the tally only climbs toward the stopping point when the sample consistently supports the outcome. The group works through both pairs until each reaches its threshold, which took thirteen ballots in this run. A parallel column shows how the same draws would look if the reported winner were wrong, drifting away from the stop point.

The takeaway is that a small random sample can confirm a wide-margin outcome with high confidence, and the tally rule is simple enough to run by hand. The presenter notes that sampling with replacement keeps the arithmetic easy, while the without-replacement version requires values that change with each draw and is better left to software.