|
|
|
FreeRepublic , LLC PO BOX 9771 FRESNO, CA 93794
|
It is in the breaking news sidebar! |
In a properly-designed system, there should be less than one undetected read error per million ballots (there may be some ballots which the system flags as unreadable; those should be handled by hand). Thus, if 100 ballots are selected at random, all 100 should match perfectly.
If discrepancies are noted between any of the physical ballots and the electronic copies, then all the paper ballots should be rescanned and compared against the original scan. All ballots which don't match should be checked to determine the cause of the mismatch. Once any problems are repaired, a sample of 200 ballots should be taken; all 200 should match perfectly. If not, continue problem analysis.
If the margin on a race is less than 3%, then using a sample of more than 100 ballots would probably be a good idea. But the odds of a 3% error rate (whether intentional or not) being undetected in a 100-ballot sample are less than 5%.
To my mind, designing a system to allow hand auditing would be much more practical than allowing only machine recounts using machines which may be just as defective or rigged as those on the original count.