Boulder County is planning to use scanning and image processing to interpret and tabulate hand-marked "fill in the box" paper ballots for the 2004 general election. CVV, the commissioners, and the elections division are trying to audit a randomly-chosen selection of the "live" ballots in the election, which would also be interpreted and tallied by hand, concurrently with the machine processing. The manual results would be compared with the machine results. If they aren't the same, one of them is wrong, and the discrepancy should be resolved via further counts and investigations.

California, for example, audits all its elections by requiring that 1 percent of all paper ballots be manually recounted,whether or not an election is contested.

In addition, there is some interest in hand-counting a "statistically valid" sample that could be compared with the overall election results. For close elections, the number of ballots that would need to be counted to get a margin of error that is comfortably smaller than the margin of victory could become very large.

Memos between the Boulder County elections manager and Hart InterCivic about the practicality of statistical sample hand-counting:

• Excerpt of a memo from Tom Halicki to Hart: in PDF format or plain text format
• Hart's response to Tom Halicki: in PDF format or plain text format

(In the above memos, "CVRs" are "cast vote records," which apparently are data records which represent the vote counting machine's interpretation of each ballot -- e.g., "ballot 1 voted for Candidate A", etc)

Jonathan Wand and some colleagues at Stanford have been working in this area.

• Jonathan Wand's election-related research