I'm hearing lots of fascinating stuff at the NIST Common Data Format Workshop, but a couple of items really struck me this morning: the contrast between a presentation by a voting system vendor, and a developer of an open source balloting device prototype. Neither of them explicitly asked a very good question about central or precinct count optical scan devices that count paper ballots: what is a ballot counting device for? One obvious answer is "to count votes." Another answer is "to find votes on paper ballots, and count the votes." But this addresses only the easy case in which a voter votes correctly on every contest, sufficiently legibly for the software to recognize every mark with very high contest. So I would say something bit less pithy: "to try to find votes on paper ballots, count the votes is possible, and inform the user or operator of cases where it isn't possible."
What about the rest of the time? The open source guy had a rather different comment: "Election officials should not be asked to capture voter intent." Fair enough! As a combo techie/activist/lobbyist, that is a fine position for him to advocate. But I'm a bit more pragmatic about what happens today, particularly in the increasingly common case of centrally tabulated paper ballots, typically vote-by-mail cases. When there are marks that are not sufficiently legible, or missing, or apparently too many marks, there is no voter there to ask to re-try. The way U.S. elections work today, officials are responsible for capturing voter intent in the cases where the software can't do the job with high confidence.
By contrast, the voting system vendor person talked about their open data format (or soon to be open; they haven't published it yet). One part that leapt out at me is the ballot-record dataset, which includes info about whether or not a ballot needs adjudication, or has been adjudicated by an election official. Well, of course I like that, because it is very similar to what we're doing in our open-source ballot counting software, including the next step that we will take, of recording everything about every ballot-adjudication case, and enabling publishing of these records for transparency.
So the vendor's remarks led to a better definition of the basic function of a ballot counting device: to take a batch of ballots, and separate it into 2 groups, those that require human interpretation, and those that software was able to count with high confidence. A whole lot follows forward from there, but that is a basic starting point, based on the realities of U.S. election practices today.