I think the big problem with obscurity is that its impact is asymmetric in the wrong direction: it inconveniences white hats a lot more than black hats.
Even if mathematically unbreakable, the implementation won't be.
This is the whole premise of defence-in-depth... delay rather than prevent.
An acceptably low break in rare using mathematically valid encryption.... Yes, fine... Given a perfect implementation.
You haven't got one of those.
I don't think it's too much to ask before adopting a given security policy that it provide some evidence that it increases security. Or should I also be gathering a collection of rocks that keep hackers away?
This applies to security and safety. There is no reason not to implement obscurity also.
One thing I can guarantee is that your implementation of one security-type will not be unbreakable.... and neither will mine.
Better to add obscurity than not.
For example, hiding the algorithm from whitehats may prevent/discourage them from hunting/reporting bugs.
Yea, this is a serious concern, I guess it depends on use cases. For sure "security by obscurity considered harmful" could be true, thats the thing people overgeneralize and fight over it when this should be weighted depending on the circumstances.
Thats the problem right there.... not obscurity.