Yes, that’s it!
Observations are put in a project at an high taxonomical rank when the computer vision does not provide enough confidence to put it at a lower rank.
The next question could be: why does my software put so many observations (possibly of lower interest, if nothing easily identifiable is shown) in a few high rank projects?
By putting observations in a project during a run of the software, it becomes easy to filter them out on the server side during the next run, so that they are never downloaded again. And, in a near future, I would like that this software has no more local cache on the disk (a cache of all observations downloaded). The idea is to put every “unknown” observation (or at least those that “Needs ID”) to one project and then to ignore it, for sparing ressources. The purpose is to be able to deliver the software to other persons that would accept to contribute to populating the projects, collaboratively. Only the iNat server would know which observations have been treated already and need not be downloaded and analysed again by the software, whoever has treated these observations and whoever is presently running the software.
For exactly the same reason (avoiding to treat twice the same observation, even when using the software collaboratively) I also use “exclusion projects” for observations that are not put in a “phylogenetic project”:
https://www.inaturalist.org/projects/exclusion-list-for-phylogenetic-projects
Once I have finished to put in these projects all observations I have in a local cache (containing ID of observation + 10 computer vision suggestions / observation), I will just delete this cache.
Initially this sofware was designed for downloading, storing and identifying the observations without ID (as presented here and here). Presently it also puts observations in projects. Soon it will only download observations, analyze them and put them in projects, then forget about them.