The role of government in forming and coordinating R&D consortia has often been cited in studies of the economic success of latecomer countries such as Korea and Japan. Most previous studies documented the government's efforts to provide funding. In our research about the government's role in determining the quality of innovation, we develop a computational model based on genetic algorithms. The two main aspects of government involvement explored in this study are 1) the timing of evaluation of participating firms in a consortium, and 2) the form that these consortia take. In terms of the timing of evaluation, we find that continuous evaluation is consistently superior to early evaluation. In addition, the effect of the form of the consortium depends on the timing of evaluation. An inverse pyramid arrangement, which emphasizes variation at the beginning of the innovation process, outperforms a pyramid-form arrangement only when evaluation is continuous. We identify the tension and reconciliation between diversity and selection as the force underlying the results of this study. We discuss these findings and their implications for how governments should balance diversity and selection when designing innovation systems.