How could you distinguish between pointless or even damaging customs and advantageous ones? Spending vast resources looking for basically anything at all which

How could you distinguish between pointless or even damaging customs and advantageous ones? Spending vast resources looking for basically anything at all which gives the in-group a statistical advantage is not going to be cost-effective – researchers could probably come up with millions of different reasons which would all have to be tested. My expectation would be that the vast majority of weird customs are not useful, and that we are interested in examples like manioc processing because they are the rare exception rather than the rule.