There are two separate issues/problems that are getting combined in this stream.
Steve Ricke has been running tests on a specific unit from Seminole. He had a checksum error occur and had the same result of the card resetting to pre-election mode and being able to reset for election mode and continue. After that one error, he has since run thousands of ballots through without a repeat of the error. The original audit report for the Seminole corrupted memory card showed that it had experienced the same error when Mickey Martin and company were recounting ballots on November 9, 2000. Still testing.
Below is the sequence of events for this error. Hope it helps.
Will continue to try with other cards and accu-votes from other counties.
Thanks Guy, - the pollworker did restart the unit and eventually put the unit back in election mode. It did not require redownloading the card. Am I missing something in your explanation to understand this?
This is an overview on what memory card checksum errors are. Exactly what causes them is a separate question.
The memory card is very simply a programmable memory device with a battery backup. The Accu-Vote accesses this memory directly. If something goes wrong when the Accu-Vote is writing new data to the memory card or if the Accu-Vote crashes (as computers have been known to do) and writes to random memory locations, then the data on the memory card may be corrupted (nasty word I know but it fits). All this means is that the data is modified in an unintentional manner. This could also happen without an Accu-Vote through static discharge or some types of radiation (i.e. old airport scanners, cosmic rays???).
There are several mechanisms that we could use to detect this. We use the simplest of these which is to treat the data as a series of numbers and store totals of sets of those numbers as separate data known as checksums. If the data has been modified without updating the checksums, then the checksums will fail to add up.
The Accu-Vote keeps three different types of checksums for three different classes of data. These are text, counters, and precinct. The text checksums cover all the titles and names that are used mostly just for printing reports. Since the text data does not affect the other operations, we check it only occasionally and we allow most operations to continue after a warning.
The counters and precinct data are considered critical and the Accu-Vote is largely inoperable when these checksums fail. We do support the option to clear the counters if only they have been affected and then counting may be restarted. However there is no way to recover from corruption of the precinct data other than to clear and re-download the memory card.
All checksums are validated upon insertion of a memory card or at power on. Thus this is the most common time to detect problems. However the counter and precinct checksums are validated every time a new ballot is scanned. If an error is detected, counting is aborted.
Now to Lana's questions. The above should answer everything other than why erroneous data managed to upload. I see two possible explanations. One is that the data was corrupted after the checksums were validated. In this case the errors would show the next time the checksums were checked. The other possibility is the miniscule chance that the erroneous data managed to add up to the correct checksum. The checksums are stored as totals ranging from 0 to 65535 so the chance of this happening are less than 60,000 to 1 just based on that. Other factors add to this to make it extremely unlikely. However in this case the card would not later show checksum errors.
So John, can you satisfy Lana's request from this? I can't without more details.
John McLaurin wrote:
Please see below and let me know what you
think. Tab, one of these issues
Please let me know what you guys think.
Sophie & Guy (you to John),