Snapshot Protocol Trip to Florida 2016 Nov
Citizens Oversight (2016-11-14) Ray Lutz
This Page: http://www.copswiki.org/Common/SnapshotProtocolFlorida2016Nov
More Info: Election Integrity
, Election Team
, Snapshot Protocol
, Snapshot Protocol
Here are my notes regarding the trip to Florida in 2016 just after the election. (Still have many pictures and videos to add to the website.)
Sarasota County Elections office
Visited on Monday, Nov 14. No activity in the office. They gave me some numbers in terms of raw counts of votes.
About 250K voters in the county, roughly evenly split between polls, early and later VBM. They let me take a couple of
pictures through the window of their equipment.
They conducted what they call a "recount" of a race for City Council seat in Seminole City. When they do
a "recount" they do not conduct any audit.
My observations about the recount procedures
- Staff was very cooperative and helpful in answering my questions, Esp. Justin Latimer, their PR mgr.
- Meeting was nicely conducted in a very professional manner.
- They had a written agenda which they followed step by step
- My camera was allowed to video record the operation through the glass windows. They used a PA system so we could hear the conduct of the meeting. This was good.
- Members of the canvassing board and staff sat around a table and did not face the public.
- Logic and accuracy test:
- 135 ballots were run through the scanner to show they counted accurately. Two reports were compared from the machine, one before and one after they scanned the ballots. After the ballots were scanned, the Canvassing Board compared the two reports and placed the test ballots into a lockable box with "L & S" stenciled on the side.
- The Canvassing Board did not actually count the test ballots nor did I witness anyone else count these. They just compared the reports.
- Since the reports were both generated in teh same format, it seems the way they generated them was to run the same ballots twice through the same machine, first pass provides the first report and second pass, the second report. Since I witnessed no review of these ballots by the board, it would be possible that the scanners made the same mistakes twice, and they just confirmed that.
- Since we did not see the ballots in the L&A test, we don't know if they represented a thorough test of the machines. Also, three machines were used in the actual counting, yet we saw only one machine used in the L&A test. 135 test ballots sounds like a very low number for such a test. we did confirm that the test did include over and under votes.
- When the ballots are scanned for under and over votes, they are distributed into separate bins using a sorting function. This function was not tested in the L&A test.
- The L&A test did not test for failure conditions, such as a paper jam. Such a jam DID occur in the rescanning process.
- The Canvassing Board sat at the table and told irrelevant stories during the testing phase and trusted staff to to the right thing. For all they knew, staff could have pre-produced a report, pulled in out of a drawer, and then photocopied it to make enough copies for the board and staff members.
- Again, the canvassing board did not compare the actual ballots to either report and they did not try to witness the L&A process.
- The Canvassing Board then signed off on the L&A test. With their level of scrutiny so low on this phase, it does not give observers much confidence they are providing adequate oversight to the rest of the process. Most significant is the lack of review of the ballots and the fact the test did not cover the sorting function nor failure modes.
- The next step is to rescan all the ballots and separate all the ballots that had under or over votes. In this race, there were three candidates and it was vote for two. (NOTE is this correct?). Thus, an under vote would occur frequently if anyone decides to vote for only one and not two candidates, which is a common strategy.
- Boxes of ballots of the precincts were wheeled into the room and workers scanned them through three DS850 scanners.
- These scanners scan the entire ballot, front and back, creating a digital image, and then process that image to extract the vote. The digital image is a "document" created in the process of tabulation and in fact, the vote is recognized from that image. The DS850 Scanners used by Pinellas County must make the digital image so the vote can be extracted.
There is an option to delete these images as they go. Using this function during an election or recount is a violation of the federal law which requires that all records be kept for a period of 22 months.
In my discussion with Justin Latimer, he explained that Pinellas County routinely violates this federal law by selecting the mode where the images are discarded and not saved.
- The Scanners sort ballots into one of three bins:
- under votes,
b. over votes, and
c. tabulated votes.
The tabulated votes are not inspected at all. The under votes and over votes are later inspected by the canvassing board.
- During the scanning process, one ballot was destroyed by the sorting portion of the scanner. The ballot had to be "remade". However, since the ballot had already been scanned and was in the process of being sorted, it is unclear if ballot had to be rescanned or not. It seems logical that that ballot would not need to be rescanned as the scanning process was complete and it was simply being sorted. After the crumpled ballot was removed from the mechanism, it was transported to the remaking room and copied to a new ballot.
Since the L&A test did not test this failure mode, and there was no review of the counts by actually counting ballots, there is a great deal of trust put into the machines and the super-simple and abbreviated L&A test.
- I and others in my group were unable to stay for the actual review of the over and under votes.
- In Florida, "recounts" are automatic if any race is within 0.5%, when a automated tabulation is performed, and if the race is within 0.25%, then a "manual" recount is performed. We are concerned, surprised and dismayed that a "recount" relies heavily on the automated scanners even if it is called a "manual" recount. Plus, if a recount occurs, then the random audit does not. Unfortunately, the features of and audit are NOT replaced by a "recount" in the way they are currently implemented.
We recommend that counties in Florida add some of the features of the audit to the recount to deal with the fact that these shortcomings exist. When a recount is performed, 1% of the precincts in that recount should not only be tabulated by machines, but also subjected to a manual tally of the ballots. That would provide the coverage of the vulnerabilities normally detected by an audit and improve the level of scrutiny in the recount process.
We would prefer that this would be applied to any and all races that are recounted, but if multiple races are subjected to a recount in one county, then at least one should be also subjected to the manual tally procedures normally in place for an audit when there is no recount required.
- In Florida, "Audits" are performed after the election is certified. Only one race is selected at random and only 1% of the precincts appropriate for that race are included.
The fact that it is after certification is partly a good thing and partly not so good. The reason we like this as opposed to prior to certification is two fold.
\ First, the results of the election are complete and thus election officials can publish the results broken down by precinct (and email batch as appropriate) and these are completely frozen and published. Thus, there can be no "fixing up" of the results before the audit to cover up any central tabulator hacks.
Secondly, the fact that it means it is after certification means it can be challenged and such a challenge does not directly affect the results. Thus, the audit is more like a recount. This is in contrast with the situation in CA where the audit is conducted prior to certification, resulting in a situation where it is nearly impossible to challenge the procedures of the audit without the county using Election Code section 13340 which says a writ of mandate can only go forward if "That issuance of the writ will not substantially interfere with the conduct of the election."
- We did not stay for the actual review of the under and over votes.
In the primary of August 30, 2016, we were disappointed that Miami-Date first published that they would do the random selection process on Monday, Sept 5. But when our volunteers attempted to go, we were informed that they had already completed the random draw, and thus it was not feasible to see it.
This time, one race was close enough to prompt an automatic recount. I did not know it was the case until I went by the office just to case it out one day before the random selection meeting. I went in just before 5pm and I asked them where the random selection was to occur the next day. They said they were already doing it. After some confusion I realized that they were conducting recount procedure. I video recorded some of the recount process from the observation area.
They were very liberal in allowing people and cameras to witness the recounting process.
Palm Beach County
I still wanted to see a random selection process for myself. The videos of these procedures left a lot to be desired when I was able to see the procedures in Citrus and Marion Counties, and Broward county. The only one in the area that was not being over-ridden by the need for a recount was Palm Beach County. I went there and hit pay dirt.
There is a separate topic for this random selection meeting.