This paper presents a level generation method for Super Mario by stitchingtogether pre-generated "scenes" that contain specific mechanics, usingmechanic-sequences from agent playthroughs as input specifications. Given asequence of mechanics, our system uses an FI-2Pop algorithm and a corpus ofscenes to perform automated level authoring. The system outputs levels thathave a similar mechanical sequence to the target mechanic sequence but with adifferent playthrough experience. We compare our system to a greedy method thatselects scenes that maximize the target mechanics. Our system is able tomaximize the number of matched mechanics while reducing emergent mechanicsusing the stitching process compared to the greedy approach.