Tetris with 0 gravity and infinite next pieces has notably been proven NP-hard several years ago. The proof (available as a PDF preprint from arXiv) appears to apply to some game similar to Tetris. But some of its underlying assumptions have been rendered invalid by the Guideline. It assumes a model of "instantaneous rotation" which is identical to BPS bounding-box rotation (i.e. SRS rotation without wallkicks) except for the behavior of I. It uses different names for five of the tetrominoes, including four that I had never seen elsewhere: O -> Square (Sq) (obvious) J -> Left Gun (LG) L -> Right Gun (RG) Z -> Left Snake (LS) S -> Right Snake (RS) The proof gives four restrictions on what it calls a "reasonable" rotation model, including the following stipulation: SRS can be made to kick up or down in this case. See the recent discussion of I-spins in the Lockjaw topic. But where the paper establishes a link between Tetris and 3-Partition, it would appear that a case with T=0 or s=0 would result in a trivial case of 3-Partition. And any case with T>0 or s>1 would result in a playfield much larger than the Guideline playfield. JZJJO will never occur under the Guideline's bag randomizer unless you have two bags 12345ZJ JO67890, and 5 is held and replaced with J. But then, the article considers only move left, move right, move down, rotate left, rotate right, and lock if landed, not hold piece. In addition, the article leaves the following important questions open: I have been doodling tesselations of bag-randomized tetrominoes that may allow me to prove that Tetris with bag randomization and SRS wall kicks can be played forever from an empty well, in much the same way that Lumines has been solved.