Results might be wrong
Like any new feature, thunks carry the normal risk of bugs
Dyalog has been improving testing:
A code coverage tool tells which code is tested and which isn't
Fuzz testing helps find errors in unusual circumstances
Lowering the "thunk threshold" allows more thorough testing
A specific problem with fptrs: simple scalars might fail to "float":
0 1 2 ≡ (1500⌶0)1 2 0 0 1 2 ≡ {⍵}¨(1500⌶0)1 2 1
Thunks are either never scalars or, like Depth (≡), always scalars