With all due respect, this is not a statistical argument. The question is, "is 20,000 trained divers and 4.5 million dives sufficient experience to expose flaws in the program?" I submit that it is. This is hardly your facetious example for a 10 minute class with zero participants.
If you want to make it statistical a question you need to come up with a number of divers and a number of dives, a
messy question at best. And then you need a number of fatalities and a number of ear injuries, neither of which are available. The DAN figures while interesting, are BS since they are based solely on passive reporting.
The statistical question becomes, "is zero occurrences in a population of 20,000 with 4.5 million trials different than say (approximately) 3000 occurrences in a population of say 10 million who have learned to dive since the beginning, who have made about 75 million dives?" And the exact answer to that question is, in reality, meaningless since the assumption of a random distribution of fatalities is bogus from the get go.
I guess what the question comes down to is, "would you rather have you loved ones trained using a methodology with a zero fatality rate and a very small minor injury rate or using a methodology that has a fatality rate in the range of one in five thousand and a minor injury rate that is several orders of magnitude greater than that, but that is a half to a third the price?"
Might I point out, in case you really want to purse the invalid statistical comparison that for a population of 20,000 divers we'd expect at least 4 fatalities.