Extended Data Fig. 3: Performance was sensitive to both instance complexity measures.

a) (left) Box plot shows the relationship between complexity defined by k and complexity defined by t under the range of values used in the task. The Spearman’s correlation between the two measures is 0.94. n = 462 instances. (right) Box plot shows the relationship between complexity defined by k and complexity defined by t under finer value intervals, ranging from 0.1 to 0.7 with a step of 0.01. The Spearman’s correlation between the two measures is 0.89. This finer scale of analysis is done simply to demonstrate that the significant correlation was not based on the values we used. n = 10⁶ instances b) Box plots show the relationship between monkey G’s (left) and monkey B’s (right) average session performance (excluding break trials) in the knapsack task with four items and complexity defined by k (p = 0.038 and 0.037, for monkeys G and B, respectively, rank sum test). Only one instance with complexity k = 2 exists for the four-item task and didn’t produce enough trials. n = 8 and n = 16 sessions for monkeys G and B respectively. c) Box plots show the average session performance (excluding break trials as a function of complexity defined by t (rho = −0.29, rho = −0.22, rho = −0.39, rho = −0.32 and p < 10⁻⁴, p = 0.036, p < 10⁻⁵, p = 0.01, for top left, bottom left, top right, bottom right, respectively, Spearman correlation). The top row restricts the analysis to instances with complexity k = 1, and the bottom row restricts the analysis to instances with complexity k = 2. Orange box plots show data for monkey G. Brown box plots show data for monkey B. n = 8 and n = 16 sessions for monkeys G and B respectively. Box plots show the median (line), quartiles (boxes), range (whiskers) and outliers (+).