Why not use smallcheck?

I see your point, but I still find that this is a too narrow view on how to treat depth. Here, you implicitly identify depth with length of the array. This would lead to an unusable implementation. For example, [Integer] would enumerate all (i.e. infinitely many) one-element lists before the 2-element lists. Clearly this is not what we want. It’s not what tinycheck implements, and I assumed until now it’s not what smallcheck implements?

For your specific example, for a threshold of 18, tinycheck finds the bug after less than 2 million (and more than 200 000) values, which takes 0.3 seconds on my machine. (1.4 seconds for a threshold of 20, still less than 2 million values.) I agree that this is probably less performant than QuickCheck, but still I find it noteable. I achieved it by just writing down the basic idea behind SmallCheck from memory in little time as a library. This says something about the power of the idea.