Emma and Tara spend an weekend at a half-empty guesthouse. One evening, while chatting over a bottle of wine in the dining room, they realise it is past 10 p.m. The guesthouse owner and the few other guests have retired to their rooms. Emma and Tara would love to continue their conversation but do not want to disturb others.

“Others would agree that we keep chatting within reasonable limits. Let’s just keep our voices down!” says Emma.

“I don’t know,” replies Tara. “In a guesthouse like this, there’s probably a rule against noise in the common areas after ten. Let’s go to sleep.”

The contractualist theory developed by Levine and colleagues helps illuminate what is happening here. Emma and Tara recognise that their interests may conflict with those of others in the guesthouse. At the same time, they are navigating an unfamiliar setting and are unable to communicate with people affected by their actions. Unsure of the right course of action, Emma engages in virtual bargaining in her mind, trying to identify a solution to this specific situation on which everyone involved would agree. Tara, by contrast, does not bargain over the particulars; instead she appeals directly to what she presumes is an existing (or merely hypothetical) late-night-silence rule – invoking an action standard rather than negotiating a bespoke solution.

What, however, can explain Tara’s preference for a rule-based solution? While the Target Article devotes substantial space to the social benefits of rules – streamlining coordination, easing communication, shaping behaviour, and making compliance visible – it says surprisingly little about how an agent decides whether the specific problem they face calls for a rule-based solution rather than engaging in virtual bargaining. Such a choice is often consequential: applying a case-specific simulated bargaining solution in a context governed by an existing social norm can lead to social disapproval, coordination failures, or even legal consequences if the norm is of a more formal kind.

In this commentary, I propose that everyday moral cognition classifies certain situations as rule-governed. In such contexts, people expect behaviour to be guided by a standing – or hypothesised – norm rather than by on-the-fly preference trading or interest guessing. Once an interaction is classified this way, agents either look outward – invest effort in discovering the actual rule – or look inward – simulate a plausible rule, for instance by using the generalisation machinery the Target Article describes.

A central agenda for future research, therefore, is to chart the situational cues that trigger the “rule-governed” classification. Two candidate cues stand out and lend themselves to laboratory and field tests:

Coordination incentives. When the cost of miscoordination is high – as in road traffic, air-traffic control, or medical triage – people intuitively expect a focal point, naturally supplied by a rule. In such domains, relying on ad hoc preference trading is reckless because every actor’s payoff hinges on seamless alignment. Experiments can vary coordination stakes (e.g., low- vs high-penalty traffic simulations) and observe whether participants shift from negotiating outcomes to searching for, or inventing, traffic – code – like rules.

Private-public gradient. In the private sphere – for example, interactions with family and friends – we usually know others’ idiosyncratic preferences and expect our conduct to be judged by the care and concern we show, not by blind rule adherence. By contrast, the public sphere is marked by anonymity, formality, and a broader audience; here personalised knowledge is scarce (second-guessing strangers’ preferences may even be frowned upon) and evaluations hinge on visible rule compliance (“no talking in libraries,” “queue here”). The more public, formal, or unfamiliar the setting, the more people default to rule-based reasoning. Researchers can operationalise “publicness” through factors such as audience size, social distance, or institutional décor, and then measure whether these manipulations push decisions from preference-sensitive bargaining to explicit or hypothesised rule following.

Identifying and testing such cues will reveal how often – and how quickly – agents flip from preference-tracking to rule-tracking procedures, thereby refining the virtual-bargaining framework.

A second frontier for the framework concerns what agents think they are doing once they recognise a context as rule-governed. The Target Article does not distinguish between actors who believe a rule already exists (and therefore try to discover it) and actors who merely conjecture what a rule could be. Intuitively, people invest more effort in learning and compliance when they treat the rule as real, and excuse themselves more readily when the rule is only hypothetical. Parsing this difference could clarify why some emergent norms harden into law whereas others remain provisional conventions – a distinction the resource-rational model is well placed to explain but has yet to explore.

The resource-rational contractualist model proposed by Levine and colleagues deftly explains how agents settle on the right course of action in interactive settings with limited information and communication. As I argued in this commentary, its explanatory reach will grow once it tackles two further questions: when do agents classify an encounter as rule-governed, and how do they treat the relevant rule – as something to be discovered or merely hypothesised? Mapping the situational cues that steer these judgements – especially high coordination incentives and the private-public gradient – and tracing their downstream effects on learning, compliance, and norm crystallisation appears to be a promising avenue for future research. In short, more empirical work is needed to establish which contexts people regard as rule-governed, and the Target Article offers a robust theoretical scaffold for that agenda.