The biggest criticism of cardinal methods is that they’ll devolve into FPTP through bullet voting. Yet, from experience in both California’s Bay Area and Australia, we know that RCV elects the candidate with the most 1st choice votes 90% of the time (https://democracyjournal.org/arguments/ranked-choice-voting-is-not-the-solution/) (https://www.fairvote.org/every_rcv_election_in_the_bay_area_so_far_has_produced_condorcet_winners), and I believe the FPTP runner-up the other 10%. This suggests that the most important purpose of a single-winner method may not be changing who wins, but rather, helping new candidates grow. On that front, cardinal systems are clearly better, as evidenced by how they consistently give 3rd parties way more votes than IRV. https://www.electionscience.org/library/approval-voting-versus-irv/ and https://rangevoting.org/NurseryEffect.html
FairVote seems happy to criticize the top-two runoff system as unsatisfactory, which is interesting because it is rare for IRV and top-two runoff to differ (at least if you assume that the delayed runoff outcome would produce the same result as the pairwise comparison between the two runoff opponents on the ranked ballots.) I have often thought that it would be useful to try to determine whether there has ever been an IRV election in the US where the winner would not have been the same under top-two runoff, because if it turned out that the answer was no, then it would suggest that some of their criticisms of top-two runoff also apply to IRV. Unfortunately, making a ‘never’ claim runs the risk of missing some jurisdiction that used IRV at one time. However, if you are correct that
that information combined with FairVote’s “no Condorcet failure” claim would rule out all elections in the Bay Area dataset as cases where IRV would have changed the outcome from top-two runoff.
Of course, there has been Condorcet failure in US top-two runoff elections in the past (e.g. the infamous “Lizard vs. Wizard” race), and in IRV elections (Burlington, which would also have been a failure in top-two runoff). I think that it is interesting that FairVote is choosing to evaluate IRV on the basis of its performance of only a specific region of the US when there have been IRV elections with publicly available ranked ballot data outside this region, and that it is suspicious that the elections in their chosen region are so conspicuously unchallenging.
I think initially all systems will either pick the FPTP winner or runner up most of the time. Third parties need time to grow and voters need to start forming complex opinions.
One of the statistics was from Australia, which has had 100 years to grow 3rd parties.