http://www.unicode.org/L2/L2021/21058-kaktovik-numerals.pdf is the accepted proposal to add the numerals to unicode; it includes samples from two computer fonts and a photograph from a kindergarten. it's a shame they're not in the bmp given that there are only 20 of them. also it cites https://www.youtube.com/watch?v=EyS6FfczH0Q which is better than the article

i agree that it's disappointing that the sciam author, amory tillinghast-raby, knew so little about math that they didn't understand that what's supposed to be universal about math isn't the system of numerals; such ignorance or malicious disregard for truth is astounding in this context

as for why it's better, if we count 0 as 3 strokes (backslash, left, slash) and a base-20 digit as 4.32 bits, the kaktovik digits average 1.18 bits per stroke, versus what I calculate as 1.11 bits per stroke for our western arabic digits (using the stroke counts [3, 1, 3, 4, 3, 4, 3, 2, 4, 3])

averaging the number of strokes required per number up to 268 (a randomly selected number) we get 6.30 strokes per number with the kaktovik numerals or 6.79 strokes per number for western arabic numerals, an 8% advantage for the kaktovik numerals

the mayan base-20 numerals are more immediately comprehensible than the kaktovik numerals but i think they are harder to write and more error-prone to read

a way that base 20 is worse is that the multiplication table is substantially more unwieldy to memorize; however, if you can overcome that, both multiplication and division become more practical. for example, numbers between 1000 and 8000 have four base-10 digits but only three base-20 digits, so multiplying two of them in the usual way in base 10 will require 16 multiplication-table lookups and summing four partial products of usually 5 digits, while doing it in base 20 requires 9 lookups and summing three partial products of usually 4 digits, about 40% less work (aside from the number of strokes required)

in the limit, representing a large number in base 10 requires about 30.1% more digits than base 20, and so about 69% more work in the standard multiplication algorithm, but beyond about 5 digits you should be using karatsuba multiplication anyway

a way in which the kaktovik numerals are worse than western arabic numerals is that you definitely wouldn't want to use them to write a check; all numbers except for 20ⁿ-1 (0, 19, 399, 7999, etc.) can be increased by adding a single extra stroke to an existing digit

the chinese (base 10) system, which has a less extreme version of this problem, has a separate set of high-security "大写" or "financial" numerals for contexts where this matters https://en.wikipedia.org/wiki/Chinese_numerals#Standard_numb...