...when I was doing it, we generally validated the ballots before counting them.

So does anyone else who has any sense. The 1991 Worldcon in Chicago (selecting the 1994 Worldcon, race between Louisville and Winnipeg) managed a "critical failure" by failing to validate votes as they were cast. The administrator rather naively thought that validation would be a trivial task "Because we'll use a computer." Most of us in the counting room, especially those of us who program computer databases for a living, were horrified when we heard this.

Even at one minute/ballot, it would have taken nearly a day and a half to actually validate 2,107 ballots. Our initial experience of using the computer for validation suggested that it was going to take even longer than one minute per ballot. We ended up begging the Registrar for the master paper copy of the database by number and splitting the validation job up into three teams. (I just realized that as it took three teams twelve hours, that one ballot/minute figure is about right, since it thus took 36 team-hours to do the validation.)

At the following day's Business Meeting, at the urging of the Committee of Tellers, the Business Meeting passed standing resolution BM-1991-1: "Resolved, That the Business Meeting recommends that all future Worldcons adopt the practice of validating site-selection ballots as they are received, rather than after site-selection balloting closes."

So far, every other Worldcon has managed to avoid this particular failure.