Firms may try to maximize profits but fail for various reasons. I use a reform to argue that the only model of the firm consistent with the data is one with misoptimization due to insufficient knowledge of the world. A reform in Israel limited prices to end with 0 as the cent digit (e.g. 2.90 but not 2.99). Since consumers are left-digit biased, specifically causing demand to discretely fall by 5%-9% at round prices, optimal pricing implies bunching at just-below prices (e.g. 2.99 pre-reform or 2.90 post) and avoiding round prices (e.g. 3.00). In fact, before the reform, supermarkets set just-below prices for 45% of prices and rarely used round prices. If price-setting before the reform was driven by the correct model of demand, firms' response to the reform would have been to update immediately according to their beliefs and avoid round prices. However, pricing after the reform is inconsistent with their pre-reform revealed beliefs, setting 20% of clearly dominated prices for almost a year. Whether firms were optimizing a wrong model or making decisions in a model-free way, their knowledge had to be partial. Partial knowledge driven by incomplete learning can explain how firms behave suboptimally in a persistent way and challenges counterfactual exercises that rely on the assumption of model-based optimization. I suggest an approach of almost-optimization to calculate counterfactuals that are more likely to contain the truth.