From LWVtopics Steve Chessin Answers Questions on VVPATs

In a paper ballot under glass, what happens if the voter makes a mistake?

- The voter notices that the paper record does not correctly reflect
  her intent.
- Instead of pressing the "Accept this as my vote" button, the voter
  presses the "Reject this as my vote and let me vote again" button.
- The electronic record is voided, the paper ballot is marked "spoiled",
  and the voter can try again.

What happens if a mistake is made with a paper ballot?

- The voter hands her ballot to a poll worker and says "this is not my
  intent".  The poll worker puts the ballot into the "spoiled ballot"
  envelope, and reactivates the card.
- The voter tries again.  If she likes the result, she puts the paper
  ballot into the ballot box, and hands the card back to the poll
  worker.

Suppose a voter doesn't verify her ballot?

A voter could skip the verification step if she chose, and just blindly
accept the paper without looking at it.  (Strictly speaking, VVPT
should be called "voter verifiable paper trail", as opposed to "voter
verified paper trail", as there is no way to force a voter to examine
the paper record.)  But the paper record, if accepted, would be
presumed to be the voter's intent, and would override the electronic
record, in case of any discrepancy.  (The analogy is with contracts.
If you sign a contract, it is presumed that you have read it, even if
you haven't.  If you tell the machine to accept your vote, or if you
put the paper ballot in the ballot box, it is presumed that you have
verified it.)

Isn't parallel testing sufficient?

Parallel testing only tests one or two machines per county.  VVPT
effectively tests every machine.  If there is a flaw or malware that
affects only a small percentage of the machines, it is more likely to
be caught with VVPT than with parallel testing.  Also, the 1% manual
check will be able to catch any discrepancy between the paper records
and the electronic counts.

Parallel testing also doesn't address the question of what to do if an
unresolvable discrepancy exists.  Do you void the election and rerun
it?

Finally, that the machines passed the test on March 2nd, 2004, doesn't
prove that they will operate properly on November 2nd, 2004.  As stated
in the consultant's report to the Secretary of State:

        The Program results provide a "snapshot" of a specific Election
        Day. Thus the value of the results is limited to the March 2,
        2004 Election Day. The Program would need to be repeated in
        future elections in order to provide the extra verification of
        accuracy that was provided for the March 2, 2004 election.

Won't the 1% manual recount catch malfunctions?

In California there is a mandated 1% manual check of
the optical scan sheets against the electronic totals.  This would
catch any malfunction of the counting software.  (I have observed this
manual count.  They do indeed count them by hand.

Electronic malware, on the other hand, doesn't need to know before the
election where the Democrats and the Republicans will appear on the
screen; it can "read" the screen at the same time that the voter does.
It can also keep track of the date and time (and I have previously
described how it can detect attempts to fool it by changing the date
and time) and operate flawlessly except during actual election use,
thwarting all attempts to expose it.