We're using a room acoustics model that captures early and late reflections based on the acoustic properties (dimensions and materials) of the room. :)
Hi Science404, yes indeed we're launching with the standard shoebox for now, but obviously we're thinking about the future too. :) Currently we calculate listener-based 1st-order reflections, optimized for performance. Once the ecosystem out there gets faster, we can explore fancier methods. ;) You can see this in EarlyReflections.js