Standard criticisms of unjustified grammaticality assignments in generative syntax are easily countered by reminders that grammaticality is a different notion from acceptability. In response, the criticisms are reformulated here in a way that interprets unjustified assignments as possible evidence of the predictive failure of current generative syntactic models, and not, as is standard, as evidence of an overly narrow evidentiary base. This is argued to invalidate appeals to the grammaticality/acceptability contrast, showing that the persistent anxieties about intuitive grammaticality assignments in generative methodology withstand the stock Chomskian defence. It is argued that if grammaticality assignments are as unreliable as critics of generativism often claim, their use conflicts with both the obvious theoretical goal of Minimalist syntax, and with its orientation as "Galilean" science.