Outcomes are finite, 0 - 8. - Poisson is out of the question, though it may prove a useful approximation.
p = 0.10, the probability of failure, is nowhere near 0.50. Normal approximation probably isn;t very good.
A radio is either good or bad. - Maybe a Binomial model?
Okay, Binomial it is.
p(failure) = p = 0.10
p(working) = 1-p = 0.90 = q
n = 8
p(at least one defective) = 1 - p(none defective) = 1 - 0.90^8
Kind of a funny question, actually. If it is known to be a 10% defective shipment, why bother to test any? Either take it or don't. It's still 10%.