MIT researchers have proved that determining whether Mario can finish a custom Super Mario Maker level is undecidable, placing the problem in the RE-Complete class. The proof comes from Erik Demaine’s algorithmic lower‑bounds class, where four students built levels using fan‑made editors that simulate logical gadgets. No algorithm can always predict Mario’s success in these constructions for any such level.

Earlier work showed Mario puzzles are at least as hard as the traveling‑salesman problem and integer factoring, situating them in PSPACE. Demaine had long argued that PSPACE was Mario’s permanent home, but the new gadget‑based reduction pushes the complexity beyond PSPACE. The counter gadget counts monsters, enabling a simulation of arbitrary computation within the game world for researchers today again.

The findings illustrate how video‑game level design can encode formal logic, offering a sandbox for complexity theory experiments. By turning Mario’s jumps and doors into truth‑value gates, the team demonstrates that even a beloved platformer can host undecidable problems. This bridges entertainment and theoretical CS, showing that game creators inadvertently craft computationally universal systems for future research and design studies.