The behaviour desired is for the observation taxon to update as described. However it is implemented - a ‘reasonable’ algorithm (according to me) or not - the CID behaves as people expect.
Best algorithm (for both Id’s) so far: https://forum.inaturalist.org/t/community-taxon-algorithm-tweaks/28583/32?u=tchakamaura
CID will not go to species, in these cases, but the decisions at least make sense at every step.