Sure, why not? I have no problem with counting machines that aren't actually used to produce the official count.
Doing manual counts only when it's close isn't good enough though, as that leaves you open to a vulnerability where the attacker simply arranges for it to not be close.
> a vulnerability where the attacker simply arranges for it to not be close.
That attack, for state and federal elections, involves falsifying multiple independent organizations' poll results, including exit polling on the day of. Otherwise the "not close" results will not be credible and they will be investigated.
Doing manual counts only when it's close isn't good enough though, as that leaves you open to a vulnerability where the attacker simply arranges for it to not be close.