Your link clearly says, in heading: "RCV for Single-Winner Offices (also known as Instant Runoff Voting / IRV)". So unless they are not actually using IRV, then the link I supplied is relevant. It's specifically this part which is broken, quoted from your link:
The candidate with the fewest votes is eliminated, and voters who picked that candidate as ‘number 1’ will have their votes count for their next choice.
So unless a candidate gets a good-enough first-vote showing to not be last, they have zero chance of winning. Even if 100% of the population picks them as their second candidate, and the remaining first-pick votes are spread over 100 other candidates, it doesn't matter they will lose without first-pick votes. And that's what results in graphs like the "square" graph in my link, where a centrist population ends up with a candidate that doesn't make sense.
The heart of the matter is that part of the vote information is ignored until other events happen. But by the time that triggering event happens, the information may no longer be used. In my example, the fact that 100% of people pick a candidate as their second choice is flat-out ignored until candidates start being eliminated. By the time that second choice is looked at, the relevant candidate is no longer in the running.
I have not seen similar multi-candidate simulations, but STV seems to have the same issue regarding ignoring information. Unpopular first vote candidates will be eliminated, even if they're extremely popular second vote candidates.
For single-winner elections, approval voting is both simpler to administer (because there's no ranking) and has better outcomes (because all information is taken into account at the beginning). However, a more complicated method would need to be used for multi-candidate races if "fairness" of representation is a target.