Judges Evaluation
The following are the collective comments of the judges on each finalist.
Punchscan
The Punchscan team provided a very complete and thorough voting system. The system did everything that it advertised, and was used in a full student election with multiple poll sites and trained poll workers who had not been part of the system development. The team innovated to meet the requirements imposed upon them by the election, and thought carefully about issues such as accessibility and mail-in voting. The Punchscan team was also quite thorough in examining the work of other teams and providing valuable critiques. Overall, the work on Punchscan was very impressive.
The most significant negative of Punchscan was the complexity of the ballot format. The form was unfamiliar to most voters. The indirection and forced random selection by voters were potential causes of concern. The physical lock to prevent chain voting, while innovative, was cumbersome and its use was inconsistent and difficult to explain to voters. Nevertheless, and despite some ballot casting errors, the Punchscan approach and the efforts of the team were commendable.
Prêt-à-Voter
The Prêt-à-Voter design was judged to be the best of the competition. The ballot form was intuitive and easy to use, and it seemed to cause little difficulty in voting. Limiting the random selection to which ballot paper to pick up seemed to keep the task manageable. The step of giving each voter a digitized scan of his or her ballot paper rather than original, and thereby forcing a real-time comparison, was cumbersome, and the requirement that voters fill in pre-marked 7-segment displays was a rather awkward way to handle OCR difficulties.
It was disappointing that the Prêt-à-Voter team could not complete their student elections, but this situation was out of their control and they should not be held accountable for it. The weak random number generation method (discovered by the Punchscan team) was a significant implementation flaw, and the decision for tellers to obtain randomness by polling other tellers seemed questionable. It would be nice to see a version that does not depend upon availability of all tellers throughout the full election.
Prime III
The emphasis of the Prime III work was on the front end, and there was good work done here, although ambient noise seemed to cause a lot of difficulties with the voice recognition system (perhaps sound absorbing materials or noise-canceling equipment would help). It was a disappointment that this system did not provide any voter verifiability, and without this, traditional security methods are the only fallback and must be taken much more seriously.
The use of a video record did not address many of the back-end tampering vulnerabilities and introduced a channel for coercion that was not adequately considered. It would be great to see the Prime III team return to the competition with a back end that could provide true verifiability, but if this is not done, a more careful and comprehensive approach to security becomes vital.
Voting Ducks
The Voting Ducks team gave themselves a large task in implementing from first principles the ideas from a paper they wrote describing a possible remote voting system. There is a lot of skepticism among the judges as to whether remote voting can really provide suitable security, and the opportunity to subsequently change votes did not seem to offer a complete solution to the problem. Nevertheless, it is a problem worthy of further research.
The usability of the Voting Duck’s implementation was quite crude and asked a lot of voters, but the multi-channel approach to security is interesting in an environment where remote voting is to be deployed.