It is desirable that data analysis tools become more autonomous in managing their computational resources, minimizing risk and cost, assessing their errors, developing new representations, and integrating with other data analysis tools. To aid in this development we introduce a design for using Meta-Reasoning in a Data Analysis Tool Allocation system that employs analogical and structural reasoning to learn from and across domains and its experience. We give a formal definition of a “Data Analysis Game” that allows the implementation of a hierarchical learning framework suitable for describing and exploring real-world analysis problems. The framework may also be used to analyze the performance of the Tool Allocation system itself, allowing it to self-optimize. If the integration of tools is performed correctly, it should allow a cost-efficient level of performance not obtainable with a single tool alone or with unsystematic use of a group of tools.