Extended Data Fig. 9: Coevolution of the number of observations M and the strictness threshold q under scoring.

a, Estimated basins of attraction after competition of all possible thresholds for M ∈ {4,6,8} and the two unconditional strategies. The mix between two moderately tolerant discriminators has the largest basin of attraction regardless of M. b, Estimated basins of attraction after competition of all possible thresholds for M = 2, with a large basin of attraction towards the ‘look twice, forgive once’ and strict discriminator mix. c, Competition between the mix of M = 2 and the mixes that evolve in larger M ∈ {4,6,8} in the presence of unconditional strategies. Such dynamics result in new mixed equilibria, which always contain ‘look twice, forgive once’ and an aggregating discriminator from larger M. Specifically, the strict discriminator with M = 2 is outcompeted by less strict discriminators but with larger M ∈ {4,6,8}. The rates of cooperation at the discriminating equilibria are all > 99.5%. For all panels, the benefit-to-cost ratio is b/c = 5, with error rates of assessment and execution α = 0.02 and ε = 0.02, respectively.