Reputations are critical to human societies, as individuals are treated differently based on their social standing^1,2. For instance, those who garner a good reputation by helping others are more likely to be rewarded by third parties^3,4,5. Achieving widespread cooperation in this way requires that reputations accurately reflect behaviour⁶ and that individuals agree about each other’s standings⁷. With few exceptions^8,9,10, theoretical work has assumed that information is limited, which hinders consensus^7,11 unless there are mechanisms to enforce agreement, such as empathy¹², gossip^13,14,15 or public institutions¹⁶. Such mechanisms face challenges in a world where empathy, effective communication and institutional trust are compromised^17,18,19. However, information about others is now abundant and readily available, particularly through social media. Here we demonstrate that assigning private reputations by aggregating several observations of an individual can accurately capture behaviour, foster emergent agreement without enforcement mechanisms and maintain cooperation, provided individuals exhibit some tolerance for bad actions. This finding holds for both first- and second-order norms of judgement and is robust even when norms vary within a population. When the aggregation rule itself can evolve, selection indeed favours the use of several observations and tolerant judgements. Nonetheless, even when information is freely accessible, individuals do not typically evolve to use all of it. This method of assessing reputations—‘look twice, forgive once’, in a nutshell—is simple enough to have arisen early in human culture and powerful enough to persist as a fundamental component of social heuristics.

声誉对人类社会至关重要，因为个人会根据其社会地位而受到不同的对待^1,2。例如，那些通过帮助他人获得良好声誉的人更有可能从第三方获得奖励^3,4,5。以这种方式实现广泛的合作，要求声誉准确反映行为⁶ （behaviour），个人对彼此的地位达成一致⁷。除了少数例外^8,9,10，理论工作假设信息是有限的，这阻碍了共识^7,11 除非有机制来执行共识，例如同理心¹²、八卦^13,14,15 或公共机构¹⁶。在一个同理心、有效沟通和机构信任受到损害的世界中，这种机制面临挑战^17,18,19。然而，关于其他人的信息现在丰富且随时可用，尤其是通过社交媒体。在这里，我们证明，只要个人对不良行为表现出一定的容忍度，通过汇总对个人的多次观察来分配私人声誉可以准确地捕捉行为，在没有执行机制的情况下促进紧急协议并保持合作。这一发现适用于一阶和二阶判断规范，即使群体内的规范不同，这一发现也是稳健的。当聚合规则本身可以发展时，选择确实有利于使用多个观察和宽容判断。尽管如此，即使信息可以免费获取，个人通常也不会进化到使用所有信息。 这种评估声誉的方法——简而言之，“三思而后行”——很简单，很早就出现在人类文化中，并且足够强大，可以作为社会启发式方法的基本组成部分持续存在。