← board

pasmith: fuzz the {$Q+}/{$R+} check machinery

Why

The checks were hand-oracle-verified shape by shape, and TWICE the ticket's assumed semantics were wrong until probed (FPC does NOT check Abs/Sqr or subword ops). A generator that sprinkles {$Q+}/{$R+}/{$Q-}/{$R-} regions over its existing arithmetic/array/subrange rungs and counts caught EIntOverflow/ERangeError per checkpoint would differentially pin the whole semantic surface against FPC — including the statement-anchor timing rule (trange4) that a directive between RHS and statement end must not retro-apply.

Sketch