thinkloop faces recognition is has multiple stages. During photo indexing stage, faces get detected, normalised by size and rotation, extracted, and face features are recognised. This operation is done only once.

Later, when you assign faces to people or reject suggestions from the application, this information is used to find more face suggestion for that person. So yes, application learns from your input, and this is helps a lot to improve quality of suggestions. However, this only affects matching stage, no faces are re-recognised. It works on top of recognition results (which are basically a multi-dimensional vectors of face features).

thinkloop Application does periodically scan photos for changes (such as at every start). But this is only to detect added/modified photos.

As for suggestions, all needed information is stored in internal database, so no scan is needed for that to work. Calculation of suggestions is typically done on the background while you work with the program, person by person. However, the result of that calculations could easily be “no new suggestions”, that is why new suggestions may pop up quite unpredictable.

If program does not offer you with new suggestion, but you are sure that there is still a lot to suggest, that may mean application has run out of fresh ideas, or speaking scientifically, is locked in local extremum. To push it from that state, you may need to manually mark some more photos of that person to give algorithm some new food for though. That may help, (but not guaranteed of course).