The machine was an ES&S iVotronic touch-screen, the same model recently subjected to a blistering Dan Rather investigative report, but what Welch witnessed does not seem explainable as a manufacturing defect or screen calibration problem like those exposed in Rather's report.
"Vote-flipping" on touch-screens has been documented before. Manufacturers claim votes show up for a different choice than that chosen by the voter sometimes, explaining that this is due to miscalibration of the computer's touch-screen. Miscalibration somehow never seems to happen when you use the airport touchscreens, hopping to "2" bags when you press "1" bag, but according to voting machine vendors it is not uncommon when casting votes. A touch on one part of the machine can register on a different part of the machine due to screen miscalibrations, but this doesn't seem to explain what Welch saw.
What Welch witnessed was votes that registered CORRECTLY when he touched the screen, switching later to a different vote choice, when he was almost finished voting the full page.
Welch was stunned to see a correctly marked vote take on a life of its own, hopping over to a different spot while he voted on other items. He called an elections worker over to show him the problem. The elections worker helped him re-vote the ballot, and both men watched as the vote registered correctly, but later spontaneously altered to shift to another ballot choice.
Even if voters take the extra time to page back through every ballot screen in the election, they may not catch the error - and even if they catch the error, both voters and poll workers may attribute it to voter error. Since the vote may be designed to change AFTER the page is turned, even paging back to "check" does not stop the vote from morphing back to an incorrect selection once the voter leaves the page.
What Welch saw was not a screen calibration problem because it registered on the screen correctly. It was not "voter error" because he literally watched the vote re-write itself to another selection, not once, but twice.
The election worker called the Wharton County elections office. Welch was astute enough to see that the suggested solution was not responsive to the real issue:
"You may continue on with this ballot if you like," said the elections worker after conferring with Wharton County elections personnel, "Or I can void this and you can start over."
This is a machine that had already demonstrated it can't be trusted. This is a machine that would fail the much-touted "Logic & Accuracy" testing purported to prove voting machines don't cheat. This is a machine that would not have passed certification tests had it performed this way for the test labs. This is a machine that has no business counting votes at all.
And because the iVotronic voting systems are centrally programmed, and the programming defines how the machine counts its votes, this is a machine that has single-handedly cast doubt on every other iVotronic voting machine in Wharton County.
"You can go back and check your vote before casting it," she pointed out, referring to the voter's ability to page back one by one to review each panel. But if the machine can alter a vote - especially if the timing is such that this happens after you have moved to a new page - what good will that do?
"We can print each vote out," she said, but Welch astutely questioned how and when votes can be printed, They aren't printed at the same time as the voter votes, and the printouts simply re-create what the computer program records, so what good is that?