[cryptonector]

On the other hand, if B-before-A is discouraging, then it may be very inefficient. It could act as a filter, but not so much for skill or any particularly useful personal attribute so much as "enjoys B-before-A well enough", which isn't particularly indicative of anything.

In general I do think that B-before-A is a very useful approach, but it cannot be the only approach. For one, it doesn't scale, not as the chain of Bs grows longer over time. We teach a lot more math to students today than 100 years ago, but almost certainly we're taking shortcuts that essentially amount to a,b,a,b -- how else can we manage the otherwise ever-growing cognitive burden of the sum of our knowledge?