(Ed: He submitted written comments which are available at: http://www.sos.ca.gov/elections/peas/kenny_mostern.pdf)
The canvass involves a time issue. Contra Costa County takes 2 weeks to do the canvass. It is also a physical space issue.
In Contra Costa County, 4% of the ballots were provisionals in the last Presidential election.
If there is a close race, we expand the tally.
The canvass period is fully used. In the 50th Congressional district race, the winner was sworn in on the floor of Congress before the Election was certified and before the hand count was completed. We are fighting for every dollar all the time. We do not have enough space to expand the 1% hand count. Money and time are issues.
You have to count the same units you published and publish the counts before the manual count. Counties often choose the precincts for the 1% count before the election canvass is finished. Certain categories of ballots are not included, depending on the status of the canvass. This leads to totals being off by several votes in each precinct compared to the final totals, and is totally non-transparent. Counties need to print or offer a DVD with the entire set of votes by precinct or other category, and explain what groups are included.
As an example, in the 2006 primary, in Alameda County, results of the manual tally were changed when they was a discrepancy. They were actually comparing to the wrong set of results which led to the discrepancy. I don't think there was any malicious intent.
There needs to be extensive training for elections officials and observers.
In San Joaquin County, when one portion of the VPAT ballot was unreadable, they reprinted the whole VPAT roll to count. None of the new VPAT was voter verified. There is no point in comparing the computer tally to a new computer printout.
There needs to be clear procedures for if the counts don't match. The recount should be expanded if they don't match, and the county should not be penalized; e.g. maybe the state can fund the expanded recount.
You shouldn't randomly select precincts until numbers are published first. Elections should not be certified if there are absurd/impossible numbers. E.g. in Ohio, some precincts/counties had more votes than registered voters, and they shouldn't be able to certify in that case. There need to be procedures to take a machine out of service if the VPAT doesn't work or isn't legible.
(Ed: he submitted written comments, which are available at: http://www.sos.ca.gov/elections/peas/public_comment_tamm.pdf)
The 1% audit is not adequate to catch fraud. No one is checking county and state totals. There are over 5000 precincts in Los Angeles. No one will look and add up 5000 numbers.
They should dump the database of totals every hour in a computer readable statement of vote and post it on the internet on election night, because candidates have pressure on election night to decide whether to concede.
It should be posted in a tab delimited XML file or in election mark language.
A candidate in a close race should be able to pick a precinct to be audited, and if they are right, they should be able to pick some more.
The process is not transparent. At Alameda County, the observation area was more than six feet from the nearest row of tables used in the manual tally, and about thirty feet from the last row. I could not see or hear anything meaningful.
There was a rumor that the VPATs were included in the tally, but no observers ever saw that. Maybe they were at the last row of tables.
In some other counties, observers can walk around and actually see what is going on, and ask questions, but not in Alameda County.
The results should be printed and/or available on a CD/DVD before the precincts are selected for the manual tally.
The Elections Code has many gaps, e.g.:
The canvass is 28 days. If it were to be extended, this would cause problems with recounts in Presidential elections. The Presidential electors must meet 41 days after the election. Even with a 28 day canvass, there is barely enough time for a recount, especially since it can not be requested before the 29th day and isn't required to start until 7 days thereafter.
One possibility would be to split the federal election canvass and especially, the manual tally of the federal election, from the same for non-federal races.
I favor tiered audits, but don't think a full 1% is always needed, e.g. for uncontested races.
We are often told the manual tally for optically scanned ballots exactly matches the machine count. After reading the papers of Prof. Douglas Jones, University of Iowa, I don't really think this is likely.
Joseph Hall in his paper said one goal of the audit is to "deter fraud", I would add to "detect fraud".