The paper considers the decision that arises in climate policy where there is expert disagreement about the correct scientific model and where a group of stakeholders needs to agree on a common policy target. Policy choice is posed as a decision under Knightian uncertainty, where decision-makers lack grounds for assigning a particular probability distribution across contending forecasting models. The decision is then framed as one of balancing two competing objectives that plausibly align with the dominant concern of stakeholders from each side in the policy debate. A decision criterion is proposed to identify options for compromise that balance these objectives in different ways. The criterion spans three standard non-Bayesian decision criteria. Policies generated by the criterion are combined with visual tools to communicate 鈥渨hat's at stake鈥?in an environmental policy decision in which stakeholders disagree both about scientific models and about the relative importance of risks to the environment versus risks to economic growth. The approach summarizes information that could be useful to policy-makers tasked with negotiating a compromise. The framework is applied to climate policy using DICE-2007 (Nordhaus 2008). The results highlight a basic asymmetry between the mistake of 鈥渄oing too little鈥?and that of 鈥渄oing too much鈥?that suggests a strong argument for avoiding the current status quo of global inaction.