Yes it does sound like a circular dependency. LOL

As far as I can tell the intent of the rule was to limit the NOL based on TI BEFORE NOL, regardless of the  difference in AGI because of it