Abstract: Nature's gadgets are implemented without being planned and therefore can utilize context-sensitive components. Thus functionality that would require extensive networks of context-free components can be elicited from a minimum of material. Proteins can serve as context-sensitive components for pattern processing applications. We here describe an evolutionary search strategy currently under investigation for its potential use in conjunction with computer controlled fluidics to evaluate the computational capabilities of proteins. Our algorithm employs evolutionary search not to seek an optimum, but to seek surprises. It directs experiments and incrementally constructs an empirical model from their outcome. Reward is given for discovering conditions that exhibit a discrepancy between the prediction of the current model and the experimental result. As unexpected observations are incorporated into the model, the reward associated with them vanishes. Results obtained so far indicate that evolutionary search is a useful paradigm for characterizing the phenomenology of context-sensitive components.