Roman-period six-sided dice are common in archaeological sites across Europe. While some dice approach true cubes, many are visibly non-cubic (i.e., asymmetric/lopsided) and favor certain rolls, especially the numbers 1 and 6. It is unclear if such dice were intentional and distinctive “types” used in specific games or activities, represent “cheaters” dice, or are simply part of a continuum of variation in die shape and configuration. To explore this issue, we examine shape distribution of 28 well-dated Roman-period dice from modern-day Netherlands. Results show that Roman die asymmetry varies in a continuous fashion from true cube to highly parallelepiped, where the long side is over 50% longer than the short side. We then conduct replication experiments to examine how naïve producers configure pips across a range of shapes. Our results show a production bias, where makers place the 6 on the largest die face, not to favor certain rolls, but due to space limitations and/or the order in which they place the pips. Overall, we interpret asymmetrical Roman dice as part of a single but highly variable artifact category, not distinct types. We argue that such extreme variation was acceptable because makers and users understood roll outcomes as the product of fate, rather than chance or probability. Conformity to a true symmetrical cube was not perceived as essential to die function, and asymmetrical forms were tolerated as simply part of the acceptable range in shape variation.