We're not talking about programming errors, but malfunctions in the hardware producing incorrect results - whatever the electromechanical equivalent of a bitfli

We're not talking about programming errors, but malfunctions in the hardware producing incorrect results - whatever the electromechanical equivalent of a bitflip is. Given the error rates described by Feynman, if they had to restart the entire computation from scratch after a single detected error they'd likely never have completed the task.