One method of miscounting has been demonstrated by Harri Hursi (links above). Namely, it entails flipping votes for candidate A to candidate B and vice versa, by inserting executable codes into the 1.94w memory cards associated with the optical-scan ballot boxes (the "HACK").
When the HACK is covered up by the STACK (i.e., only precincts where the ballot boxes were stuffed by colluding poll workers are "randomly selected" for hand recounting), we become victims of a scam that John Brakey has termed the "HACK and STACK."
That is, a hand recount of the HCPBs in a STACKED precinct would be found to agree with the official ballot tally even though the poll workers had shuffled ballots in and out in order to skew totals away from the way the voters actually voted.
On the other hand, the remaining, NON-STACKED precincts using optical-scan ballot boxes with 1.94w memory cards are vulnerable to HACKING, which could be adjusted to skew the official tallies to approximately the same degree as the STACKED ones. But big the difference is that any hand recount of a HACKED-but-NOT STACKED precinct would instantly reveal the actual MISCOUNT.
With 25 million voters voting on optical-scan machines in 2004, the HACK and STACK alone could have been sufficient to steal the election -- despite the fact that voter-marked HCPBs were employed. If only a few percent of the precincts had only been truly RANDOMLY SELECTED for hand recounts, the HACK would have been detected. Then we would now be talking about a conspiracy to steal the election as a PROVEN FACT instead of denigrating election-integrity researchers as "conspiracy theorists."
Moral: As long as optical-scanners are with us, we must assure TRULY RANDOM RECOUNTS.