At the bottom of the article, John mentions REX Computing, which I am the founder/CEO of. John has been a friend and advisor, and we have played around with minimal unum implementations, but haven't been able to commit the resources to a full one until after we tapeout our first chip in a little over a month (which will be using IEEE float). We also funded "dnautics" (a HN user who also commented here) to do a soft implementation of unum 1.0 in Julia.
Besides using only base 10 representation, 1) all numbers are exact even if there are multiple representations. 2) because of 1), .1 + .2 = .3, eg, arithmetic follows the same rules you learned in school for 4 function arithmetic. 3) It supports upto 56-bit precise integer math without loss of precision. Look around YouTube for crockford talks about it. Usually they're JavaScript related. Or check dec64.com
