Monday, August 27, 2007

Verifying vote counts

Even with a great set of written vote interpretation procedures, votes counted on the current crop of optical scan equipment are not verifiable.

Three of the issues are:

1. The equipment does not produce a verifiable record of how it INTERPRETED each vote on each ballot. Instead, the equipment collects votes from many ballots and offers a “batch” total as proof of proper interpretation. We do not trust this method, and prefer the ability to perform vote by vote verification.

2. The equipment is controlled by software which is itself “controlled” by parameters and option switches that control the “rules” used to interpret votes. There is no way to verify which “rules” were used to interpret each specific vote. We do not trust this unverifiable method, and prefer that there be a way to know what rules were used to interpret each specific vote.

3. The equipment does not produce a file by batch that includes for each specific ballot page, the VOTES that were interpreted by the system. Without this computer file, there is o way to independently verify that the COUNTS have been calculated correctly. Obviously, we prefer that such a file be required.

We’d prefer optical scan equipment to (1) endorse each ballot page with a unique identifier at the time that the ballot page is first scanned, (2) create an exportable file of scanned ballot images as the ballots are scanned, (3) create an exportable file of the votes interpreted from each ballot page, and (4) create and exportable file of detailed vote count totals including over, under, and ambiguous votes.

In this way, the public can independently compare individual votes (from a particular scanned ballot image) against the interpreted votes (from the corresponding interpreted votes record). Also, the file of interpreted votes can be imported into a program, such as Microsoft Access. This way the results can be independently calculated and compared against the official results.

[What is meant by verifiable?]

Good question. Hope this is a useful answer.

1. To verify, one must have access to a specification of correct (expected) behaviour, transparent access to an indelible record of actual behaviour, and the ability to detect every difference between expected and actual. In voting we have several discrete steps: issue ballot, mark votes on ballot, interpret marks, and count votes.

2. In voting, only the voter can verify that votes are actually recorded as intended (expected), and this must be done before the ballot is cast. It must be physically impossible for any other person to make this determination. It must be physically impossible for the voter to make this determination after the ballot is cast.

3. Before voting, there must be a set of verifiable specifications that result in the particular ballot issued to a voter. These specifications must include the specification of districts, voter eligibility, and ballot style. If an ineligible person is permitted to vote, this must be detectable. If an eligible person is given the wrong ballot style, this must be detectable.

4. After the ballot is cast, each of the following discrete steps must be verifiable:

a. Batch control - it must be verifiable that all and only the ballots cast in a batch are counted. This is a verification of ballots, not of votes.

b. It must be verifiable that each individual vote is interpreted according to (vote interpretation) specifications. Because some votes can be recorded ambiguously, it must be verifiable that the interpretation of ambiguously recorded votes is consistent. In a hand counting situation, the ballots for each contest can (should) be sorted into stacks, one for each contestant, and one each for over, under, and ambiguous. In the case of optical scan count, each interpreted vote should produce a permanent record that includes the source of the vote, as well as the interpretation of the vote itself. This way, it is possible to statistically sample items, and go back to the original materials to compare the marks against the interpretation.If the scanned ballot images are recorded, then the file of scanned ballot images and the file of interpreted votes can be published on the Internet for independent verification.

c. It must be verifiable that the totals for each contest be correct, and distributed according to specification (eg. into precincts, absentee votes, districts, etc.) This cannot be done by statistical sampling. Count verification requires that all interpreted votes be independently tallied. [If original tallies are reported by batch, then individual batches can be verified, however, the process of selecting representative batches is very complex and may not result in sufficient confidence.In the case of hand counted paper ballots, the ballots in each stack can be independently counted by different teams and the results compared by an independent judge.In the case of optical scan counting, the data file containing the votes can be published on the Internet for the world to use for independent verification.

Wednesday, August 22, 2007

Not-so-secret ballots. It's worse than reported.

C/NET has published a Story: E-voting predicament: Not-so-secret ballots

The secret ballot issues raised by Moyer and Cropcho won't be solved until the toilet-paper roles used to sequentially record voter activity are eliminated.

As a Colorado poll watcher, I have the right to record the names of people who vote. By observing the sequence in which they use a specific DRE, I know the sequence of their (supposed) votes recorded on the VVPAT. Access to the roll means access to their "ballot". The canvass board and election officials have legal access to the roll.

NO!, I do not trust the officials. Not because they are evil, but because the protection of a secret ballot is sacrosanct. If any pathway to retrieving a specific voters ballot exists, it might be used: (1) by the court, or (2) for political purposes by a partisan official, or (3) to create a "threat of disclosure" needed by vote-buyers and voter-intimidators to suggest that they can know a voter's selections.

Furthermore, HART Intercivic suffers not only the problem described above, but also uniquely identifies every PAPER and VVPAT ballot with a unique, NON-REMOVABLE, serial number and barcode. Voters can make a record of this serial number on their ballot and use it to later identify their specific ballot. Consequently, the market for vote-selling is facilitated and the opportunity for voter intimidation is supported.

The arrogance of vendors who trample on our right to use a secret ballot must be punished by immediately forcing them to meet our requirements for "privately voted anonymous ballots".