CORRECT METHOD TO RECOUNT

of the VVPAT of A DRE

for the 1% Manual Tally

By Judy Alter, Director of Protect California Ballots

1. Random selection of the machine or machines to be recounted. County election officials had prepared a list of 134 TSX Diebold machines (with their serial numbers) that LA County used in early voting. I was given permission to make the random selection of the machine (1% of 134 is one). I used a standard table of random numbers in a research methods textbook I had brought with me. As dictated by the standard manner of random selection, I declared that I would choose from the right side of the rows of ten numbers. With my eyes closed I used the eraser-end of a pencil to make my selection. I had to try three times before the pencil eraser landed on a number between 1 and 134.
   
   
2. Getting the machine. Instead of getting the machine, #63, from the secured storage location of the voting machines (using the serial number to identify it), the election officials skipped that crucial step and only went to the MTS room on the 3rd floor of LA Registrar-County Clerk Headquarters in Norwalk and to the storage drawer where the VVPAT roll and memory card for that machine were stored. Not getting the actual voting machine, #63, was the first and major way the election officials incorrectly performed the recount of the VVPAT.
   
   
3. Printing out of the tabulated results. Done correctly, the officials should have plugged in the actual machine, #63, and printed out the tabulated results that are stored on the non-volatile memory in the machine. They should have compared this print-out to a second print-out that they generated by using the memory card for #63. These two print-outs must match exactly. If not, then it is possible that something could have happened to the results stored on the memory card since, in violation of HAVA guidelines, these Diebold memory cards use interpretable code instead on non-write code. The election officials in LA County in June 2006 simply put the memory card from machine #63 (I compared the serial number of the machine to the one on the memory card) into a Diebold TSX machine that they had on the table in the small conference room where the VVPAT recount was conducted. The official put the memory card into that machine and printed out the tabulated aggregated results for the entire ten days of early voting when these TSX machines were used in LA County. Another potential violation of proper procedure is that the LA County officials choose not to print out daily tally tapes every evening when the polls close during the ten-day period of early voting. They, therefore, do not keep any record of the daily totals during early voting. The two gentlemen supervising this VVPAT recount, Ed Bennett and Mike Petracello, claimed that the TSX machines do not contain an internal printer. When the technician put the memory card from machine #63 into the TSX voting machine on the table and generated the paper print-out, several yards long, of the tabulated results from the internal printer within the machine, I did not point out to these two gentlemen, in charge of the technical staff, that they were completely wrong about the TSX not having an internal printer. Together, we had just witnessed its use.
   
   
4. Readying the tally sheets to compare to the VVPAT. Two staff members cut the long tally tape (printed from the stored results on the memory card) into sections by contest and candidate and attached these tallies to separate tally sheets for us to compare to the results of the VVPAT recount.
   
   
5. Opening the VVPAT roll and counting the total ballots cast. A total of 73 voters used that TSX machine. In counting the total ballots cast on the VVPAT roll, we found that the printer jammed 6 times, an 8% failure rate. Eight percent of the voters were unable to verify that the vote they cast on the electronic voting machine matched the voter verified paper audit trail because they could not read their individual “paper trail.” Also we, who were recounting this paper roll, could not verify that this total of 73 ballots cast was accurate since the election officials used the same source for the tabulated results and the VVPAT, namely the memory card. By default, it matched. To repeat, the election officials should have compared the tabulated totals from two sources: the internal memory of the actual machine and the results on the memory card. Even at this point in the recount, the election officials were not conducting the recount correctly.
   
   
6. Selected contests recounted. The election officials decided only to count three contests. They did not know that the election code had been undated to mandate that they recount the entire ballot just as the 1% manual tally law requires for election-day ballots. As we read out the votes for governor from the VVPAT roll, we could not proceed because of the 6 unreadable ballots. After we had read out all the votes from the ballots we could read, the man in charge decided to go print out copies of each ballot on 8 by 11 single sheets of paper from the VVPAT memory card so we could find and count the votes from the 6 unreadable ballots. Next we figured out how to determine which of the 73 ballots had jammed. After studying the identification numbers on the VVPAT ballots, I found a number that distinguished the ballots one from the other. By the process of elimination, we determined which of the 73 print-outs were the jammed ballots. We then found that our recount matched the printed out totals. When we proceeded to count the next contest, we had to count it three times until we got it to match the tabulated results. If we had cut the VVPAT roll into separate ballots we could have sorted by candidate and then, much more easily, counted the results. Instead we had to rewind the paper roll several times and begin again. Instead of just being an observer, I was permitted to read out the results because the election official was tired. The roll also included 2 spoiled ballots. These spoiled ballots on the roll confused the counting because the word “spoiled” does not show on the roll until the end of the ballot. The election officials chose not to cross out those 2 spoiled ballots on the VVPAT roll; they, therefore, also caused the miscount during the process.

In sum, recounting the VVPAT roll produced challenges:

• The printer jams on some of the ballots.
• The jammed ballots cannot be either verified by the voter or easily recounted from the paper roll itself.
• The existence of spoiled ballots only identified as “spoiled” at the end,
• The procedure could be adjusted by cutting the paper rolls. This cutting up of the roll introduces another level of needed security because these small and light-weight pieces of paper could be easily lost or mislaid.

Correct procedures:

• Use the original machine from which the randomly selected memory card and VVPAT were used,
• Generate a tabulation tape from the non-volatile memory in that machine on which the VVPAT paper roll was generated,
• Compare the internal memory tape to the results generated from the memory card on which the VVPAT results are stored.
• Count all the contests in that election as required by law.