First, we assessed the overall size of the action space by counting the number of unique actions as a function of inventory size during the final ten generations. As expected, individuals with semantic knowledge or social learning explored fewer unique combinations, avoiding options that did not make sense (Fig. 2D). Second, we computed the Shannon entropy of action distributions within each inventory state. Higher entropy indicates more random exploration, while lower entropy reflects more focused, targeted search. As expected, entropy was markedly lower in populations with semantic knowledge (Fig. 2E), indicating that individuals explored fewer, more targeted combinations than individuals without such knowledge. Entropy was also lower in populations with the capacity for social learning, consistent with the idea that social learning reduces individual exploration (45). These results confirm that both semantic knowledge and social learning restrict the action space, making innovation more efficient.