2. why depth and not size (= total number of constructors)?
That seems harder to generate terms compositionally. To create all terms of depth n+1 you just glue together all terms of depth n, but to create terms of size n+1 you need to glue 1 with n, 2 with n-1 etc.
So? One would fear that this is inefficient because then these series (of smaller sizes) would have to be generated several times. But no, instead of class Serial a where series :: Int -> [a] one could "cache" these into series' :: [[a]] such that series k == series' !! k actually, series k == [ 0 .. k ] >>= ( series' !! ) because I'd want series' do be disjoint (each object has only one size). That should be doable by a simple rewrite of (some parts) of the library. I'm going to find a clever student to whom I could assign this ... Are there any standard test cases (i.e., they work nicely with the original Smallcheck and must continue to work with any modified version)? Best - Johannes.