Hmmmmm, so many points to be made here but let me start off saying that the goal, IMHO, should be to avoid false positives without excluding any true positives. I hope we can all agree to that. Next, I think it is reasonable to see that we have a grey area in that 3c and closer are not eligible for triangulation but 3c and closer below a certain level are not eligible for confirmation without triangulation. OUCH!
With that said, let me make some observations in no particular order:
1) The image above might have the top three categories closer than 4th cousin but the requirement is to be 3c or closer so we can exclude any 3c1r relationships. The total confidence for 3c or closer then, would be the sum of the1st, 2nd, and 4th entries or about 76%.
2) Since 3C and closer matches allow one to confirm the MRCA (a single person for half cousins and both for full cousins), I think for full cousins the number of matching segments needs to be some minimum. Obviously, it cannot be lower that two and with only two, there is a 50/50 chance that they both come from the same person. If we can agree on an acceptable probability that they are from two separate persons, we can mathematically derive the number of segments needed to meet that probability I should think.
3) For yDNA, we expect at least 90% of the STRs match so possibly we should look at using at least a 90% confidence that a match is 3C or closer.
4) The same DNA might register as beyond 3C on Ancestry but by downloading the results and importing to other platforms (i.e. GEDmatch, MyHeritage, etc.) one can often see them at 3C or closer. In some cases, a prediction is not even made rather one needs to take the segments and sizes and use a source such as ISOGG to help determine the probability for 3C or closer.
4) The major reason for the above is a filtering step made by Ancestry suing Timber (see: https://blogs.ancestry.com/techroots/filtering-dna-matches-at-ancestrydna-with-timber/
5) If WT uses minimum numbers of segments and minimum sizes to establish whether a match is good for confirmation, there would be quite a challenge dealing with the various sizes used by testing companies especially in dealing with short segments.
I could go on but the challenge is to decide not only what constitutes a reasonably match but also how would that be dealt with by the various testing providers and GEDmatch. A second challenge is what to do with matches at the 3C or closer level that do not meet the guidelines as triangulation is not an option under current rules (I would say that that three 3C matches could be a good triangulation or even 2 3C matches and a 4C, but maybe that's just me).