The problem:
Language resource can have several t5memory memories assigned, but at the same time segment ID is not unique among all assigned memories and can have duplicates because from t5memory perspective memories are not connected. This leads to the following: if there are duplicates in the list only the first dupplicated segment is loaded into the grid. Thus not all segments are actually visible which leads to different total amount of segments calculated and actually displayed in the grid.

Solution:
Need to make segment id to be unique for all found segments. Possible solution is to change id to e.g. UUID and change backend accordingly.