The adaptive immune system of vertebrates is able to learn from past experiences to better suit an unforeseen future. This is made possible by a diverse and dynamic repertoire of lymphocytes expressing unique antigen receptors and capable of rapid evolution within an individual. However, naturally occurring immune responses exhibit limits in efficacy, speed and capacity to adapt to new challenges. In this talk, I will present conceptual and computational frameworks to (1) explore functional impact of non-equilibrium antigen recognition and (2) identify conditions under which natural selection acting local in time can find adaptable solutions favorable in the long run – in light of viral-immune coevolution. Using these examples, I show that immune cells can exploit physical constraints to extract signal and expand their adaptive potential. Considering information propagation from receptor to repertoire and back lets us rethink the logic behind proofreading, in a dynamical system that demands precision and balance while facing perpetual disturbances.