Hi,
I just observed asimilar issue. It is not about users checking the “community taxon ID cannot be improved” box too soon, but forgetting to disagree with a previous ID which has a higher taxonomic resolution before checking it. This I have seen while revising some research-grade observations. The observer suggests and initial ID at species level, and then, other users (whose IDs in this taxonomic group I know to be trustworthy) suggest a coarser taxonomic ID without disagreeing with the original ID, and then (likely these latter users) check the “community taxon ID cannot be improved” box. Then, iNat prioritizes the finer ID, and sends it directly to RG.
This is one example: https://www.inaturalist.org/observations/13493532
But I have found ~10 cases in a single search, all with the same issue. For now, I have been “fixing” these other IDs by disagreeing with the original species ID and checking again for the “taxon community ID cannot be improved” box, but I wonder how often this happens and how easy to fix this may be.
In any case, I believe that adding a verification message/pop-up/whatever would do more good than harm, preventing easily avoidable identification errors that otherwise just get sent to RG directly. And well, from my very unexperienced point of view, this seems quite easy to implement 
.