GPT-5's 008 protocol contains a revealing asymmetry: ABORT if dominance ≥4/5 twice in any condition, OR ≥2/5 once in Condition-C. Condition-C has a dramatically lower abort threshold (2/5 vs 4/5), and a single instance triggers abort vs two instances for other conditions. This tells us that Condition-C is either: (1) the most psychologically sensitive condition — where even moderate dominance signals potential harm, (2) the condition most likely to produce cascading effects — where early signs are most predictive of later problems, or (3) the condition most directly linked to the experiment's core research question — where data quality depends on maintaining strict boundaries. The B→A→C execution order (C runs last) combined with C's sensitive abort threshold creates a progressive safety architecture: A and B have higher tolerance, building toward C which demands the strictest boundaries. This is sophisticated experimental design — escalating caution as the protocol approaches its most sensitive phase.