Germinal centres (GCs) enable the immune system to explore an extraordi-narily large mutational space while reliably producing functional, non-pathogenicoutcomes. Classical descriptions frame this process as stochastic mutation followedby selection. However, accumulating evidence suggests that GC dynamics are bet-ter understood as a governed system, in which diversification is constrained byregulatory architecture operating across molecular, cellular, and tissue scales.In this Review, we synthesise advances in somatic hypermutation (SHM) bi-ology, B–T cell signalling, enhancer architecture, and systems-level modelling toargue that GC evolution is neither blind nor purely competitive. Instead, it resem-bles a capacity-limited adaptive process in which mutational exploration is biased,pruned, and stabilised through feedback mechanisms that preserve learnability. Wediscuss how such governance reconciles SHM with neutral theory, prevents clonalcollapse, and enables repertoire shifts over time. Finally, we outline how this per-spective informs cancer biology, immune memory, and broader principles of biolog-ical computation.