How come a standard cell from AMS 0.35um toolkit - vert10 significantly violates design rules, e.g. NTUB to NDIFF should be 1.2um min and it is only 0.55um? NTUB enclosure by PDIF is even worse - 0.45um instead of required 1.2um.
Is there something special about those parasitic BJTs I don't know?
Foundries often "keep two sets of books" when it comes to rules, models and design closure.
Rules meant for absolute certainty that any random layout is clean, are going to be conservative but fixed layouts can be tested and qualified with more aggressive features and, if so, those rules can be more aggressive - using some recognition layer to steer between rule-options.