Bottom-up energy system models rely on cost optimization to produce energy scenarios that inform policy analyses, debates and decisions. This paper reviews the rationale for the use of cost optimization and questions whether cost-optimal scenarios are adequate proxies of the real-world energy transition. Evidence from ex-post modeling shows that cost optimization does not approximate the real-world UK electricity system transition in 1990-2014. The deviation in cumulative total system costs from the cost optimal scenario in 1990-2014 is equal to 9-23% under various technology, cost, demand, and discount rate assumptions. In fact, cost-optimal scenarios are shown to gloss over a large share of uncertainty that arises due to deviations from cost optimality. Exploration of large numbers of near-optimal scenarios under parametric uncertainty can give indication of the bounds or envelope of predictability of the real world transition. Concrete suggestions are then made how to improve bottom-up energy system models to better deal with the vast uncertainty around the future energy transition. The paper closes with a reflective discussion on the tension between predictive and exploratory use of energy system models. (C) 2016 Elsevier Ltd. All rights reserved.