The Synthesizability Gap Has Moved: Route Existence Is Not Chemistry

Abstract

Synthesis-constrained generative models now guarantee that every proposed molecule carries a route to purchasable building blocks, and recent campaigns have carried such designs to synthesized, biologically active compounds. I argue that this genuine success has relocated the synthesizability gap rather than closed it: a route a planner certifies is not yet chemistry a bench reproduces. Three deficits persist─stereochemical settability, chemoselectivity and functional-group tolerance, and a novelty tax that returns constrained generators to flat, amide- and biaryl-rich space─and none is a search problem. I propose that molecular design needs for synthesis what PoseBusters supplied for docking: a cheap, standardized chemistry-validity layer that scores routes rather than molecules and reports executed make-rate as the headline result.