Chater and Christiansen (C&C) offer a compelling account of how directed but myopic innovations to social rules accumulate into complex cultural systems through spontaneous order. Their “sighted watchmaker” metaphor is a welcome corrective to the selectionist orthodoxy in cultural evolution (Mesoudi, Reference Mesoudi2021), and their emphasis on coordination as the engine of cultural complexity is theoretically productive. I wish to extend their framework, however, by arguing that the very properties that make social tinkering so powerful, namely its directedness, its coordinative character, and its precedent-setting logic, also make it a reliable engine for generating and sustaining maladaptive cultural practices.

C&C emphasize that social tinkering is directed: people intelligently solve the coordination problem of the moment. This is true, but it is worth distinguishing between forward-looking coordination (e.g., jointly deciding how to divide labor on a hunt) and retrospective coordination (e.g., jointly deciding why a child has fallen ill and what must be done). Much social tinkering, especially in small-scale societies facing frequent misfortune, is fundamentally remedial and backward-looking (Murdock, Reference Murdock1980). It is oriented toward constructing plausible causal narratives for already-salient failures: illness, crop loss, or unexplained infant death (Hong, Reference Hong2024, Hong, Reference Hong2026; Horton, Reference Horton1967). The intelligence that C&C rightly attribute to human tinkerers is, in these contexts, directed not at discovering reliable causal mechanisms but at producing socially acceptable explanations and coordinated responses to distress. This is still “directed innovation” in C&C’s sense – purposeful, creative, and cognitively rich – but its direction is set by social accountability rather than empirical feedback from the natural world. My ethnographic work among the Nuosu of southwest China illustrates this dynamic (Hong, Reference Hong2022; Hong & Henrich, Reference Hong and Henrich2025). When illness strikes, Nuosu families engage in precisely the kind of directed, joint tinkering that C&C describe: a shaman (bimo) is consulted, a diagnosis is negotiated regarding which ghost is responsible, and a coordinated healing ritual is organized. This process is local, directed, and coordinative; it sets precedents that are reused and transmitted across generations; and it is, in material terms, instrumentally ineffective. Yet the system persists and grows more elaborate over time, because the machinery C&C identify operates just as powerfully where causal feedback is poor as where it is rich.

C&C’s insight that social tinkering requires joint understanding is precisely what makes maladaptive practices so culturally stable. They write that a communicative or social convention “is only effective if it is shared by the relevant participants in an interaction.” The same holds for a divinatory diagnosis or a taboo system: it only “works” socially if the community jointly accepts the causal narrative and coordinates their behavior accordingly. This creates powerful pressures toward consensus and conformity, pressures that can override individual skepticism (Boyer, Reference Boyer2020; Hong, Reference Hong2025b). Collective joint plans, then, need not be limited to mutually beneficial economic exchange or conversational alignment. They can also take the form of collective worship of local deities, the enforcement of (scientifically unfounded) food prohibitions, or the prosecution of suspected witches, all of which coordinate attention, blame, and sanctioning, and thereby “work” socially even when they fail to improve the non-social world.

This connects to a further implication that the target article underdevelops. C&C stress that cumulative tinkering generates systems of increasing complexity, and they briefly acknowledge that spontaneous order need not always lead to “progress.” Yet they do not pursue this point, and the underlying problem runs deeper than their contemporary examples suggest. In many traditional societies, taboo and ritual systems become elaborate precisely because misfortunes invite repeated retrospective causal attribution. When misfortune recurs, communities rarely discard prior rules as failed interventions. Instead, they add further prohibitions or auxiliary rituals to preserve the core presumption that misfortune has a specifiable supernatural cause, while shifting blame onto new antecedent transgressions (Hong, Reference Hong2024). This is cumulative tinkering par excellence: each round builds on prior rounds, producing dense, interlocking webs of prescriptions that are culturally stable and impressively complex, yet maladaptive in material terms. Just as the common law accumulates precedents that constrain future legal reasoning, ritual systems accumulate diagnostic precedents that constrain future causal reasoning about misfortune, generating a self-reinforcing interpretive framework that deepens with each new case.

These considerations bear directly on C&C’s six-stage model, which traces a progressive trajectory from individual tinkering to language-mediated science. The ethnographic record suggests, however, that Stages 3 through 5 can just as readily produce elaborate supernatural belief systems as beneficial institutions. The Nuosu bimo tradition involves a literate priesthood, transmitted texts, linguistically formulated diagnostic rules, and codified ritual procedures (Hong, Reference Hong2022). It exemplifies Stage 5 tinkering over explicit representations in every respect that C&C describe, yet its content concerns ghost taxonomy and exorcist protocols. Similarly, vast systems of astrological reasoning and calendrical divination worldwide represent centuries of cumulative, language-mediated tinkering over shared representations that do not track causal reality (Akasoy, Burnett, & Ronit, Reference Akasoy, Burnett and Ronit2008; Brown, Reference Brown2017). The machinery of cultural accumulation, it seems, is domain-general: it does not distinguish between representations that reliably predict the world and those that merely provide cognitively and socially compelling narratives.

None of this undermines C&C’s framework. Recognizing that the same machinery produces both beneficial and maladaptive complexity would strengthen it, while connecting it to work on cognitive anthropology (Boyer, Reference Boyer2001) and the cultural evolution of supernatural beliefs (Hong, Reference Hong2025a; Norenzayan et al., Reference Norenzayan, Shariff, Gervais, Willard, McNamara, Slingerland and Henrich2016). The challenge is to specify when social tinkering tends toward reliable causal knowledge versus self-reinforcing error. One promising direction concerns the quality of empirical feedback: where outcomes are prompt, unambiguous, and publicly observable (as in tool-making or navigation), tinkering can converge on effective solutions; where outcomes are delayed, ambiguous, or confounded by regression to the mean (as in healing or weather control), it may converge on elaborate but ineffective ritual complexes. Accommodating this distinction would yield a richer theory of how human societies arrive at both their greatest achievements and their most persistent errors through fundamentally the same process of directed, coordinated, and cumulative social tinkering.