In this opinion piece, I present four somewhat controversial suggestions for the design of futuretreebanks: a) Treebanks should be based on adversarial samples, rather than pseudorepresentativesamples. b) Treebanks should include multiple splits of the data, rather than justa single split, as in most treebanks today. c) They should include multiple annotations of eachsentence, whenever possible, instead of adjudicated annotations. d) There is no real motivationfor adhering to a notion of well-formedness, since we now have parsers based on deep learningthat generalize easily and perform well on any type of graphs, and treebanks therefore do not haveto limit themselves to trees or directed acyclic graphs.