Overlapping updates may transmit conflicting cultural rules despite accurate individual recall

A community can remember what everyone said and still end up using rules that do not fit together. The unexpected move is to locate the problem in how separate changes become an accepted public version: two changes that each work with an earlier version may conflict when accepted together. This pipeline generated that explanation as a proposal; the supplied studies have not measured the predicted transmission of conflicting rules under overlapping updates.

1. 01
   Master questionstep 01 of 04

   Cultural transmission, the passing of information and practices between people, needs explanations that distinguish how material spreads, changes, gets adopted and persists. The research goal is to identify approximately five genuinely new, testable explanation families, check whether they already exist under other names, and compare affordable first experiments with the stronger tests needed for general claims. It includes changes introduced by recommendation systems and generative artificial intelligence, software that produces new content.

   Rests on: The goal itself defines the subject as the transmission, transformation, competition and persistence of cultural information. It requires competing explanations, measurable outcomes, controlled changes that can separate causes, and observations that would disprove a proposed explanation; novelty and plausibility alone do not satisfy that request.

   Stated in the chain
2. 02
   Goal pillarstep 02 of 04

   Approximately five proposed explanation families need distinct identities and a clear account of what evidence supports them. A causal family, a group of explanations that require the same independently necessary cause-and-effect relationship, must be distinguished from another name for an existing explanation.

   Rests on: The master goal explicitly requests approximately five families, comparison with established theories, and separation of new conjectures from established findings. This stage selects the identity and evidence part of that larger agenda; its title supplies no finding about how many families actually exist.

   Stated in the chain
3. 03
   Gap questionstep 03 of 04

   The number of distinct explanations becomes a question about predictive necessity: how many remain needed after equivalent explanations are merged and each retained causal relationship is removed in an experiment? Held-out predictions, predictions for experiments not used to adjust an explanation, are the proposed way to judge whether a merger loses something necessary.

   Rests on: The preceding focus on distinct identities and evidence motivates checking whether differently named families actually do different explanatory work. The master goal already calls for competing explanations, controlled changes and decisive tests. This question makes those requirements into a proposed counting procedure; it does not report that the procedure has found a particular count.

   Stated in the chain
4. 04
   Hypothesisstep 04 of 04

   Overlapping changes to a community's shared meanings may create a conflict that survives accurate memory for each individual rule. Two people can each read an earlier public version and make a change that works with it, while their changes violate a semantic constraint, a requirement that meanings and interpretation rules remain jointly compatible, when accepted together. Later people may then inherit the conflicting accepted version. The proposed set contains four causal families. Source-conditioned reconstruction means rebuilding a message from a particular available source; variant selection means choosing and reproducing one available version; partner-to-population inference means keeping track of a particular partner's meanings while also learning what unfamiliar members of the wider group are likely to mean. The fourth candidate is concurrent-update reconciliation, the process by which overlapping proposed changes become one accepted public version. The fourth family is proposed to be necessary because knowing each partner's meanings does not by itself specify which incompatible changes become jointly authoritative. Conversely, an orderly public record does not by itself specify how a convention transfers from a familiar partner to someone new. Four counts proposed causal families, not people, messages, computers or fitted quantities; if established combinations predict the effects, the count must shrink.S1S2S3S4S5S8

   Rests on: The gap question asks which causal relationships survive attempts to merge or remove them. The endpoint supplies a proposed answer with a specific extra relationship: the next accepted public version depends jointly on the version each writer read and on which changes overlap. Its matched alternative accepts changes in sequence. The proposed test separates overlapping from sequential changes while also testing whether resetting partner history removes the familiar-to-new-partner difference. These are stated grounds for a proposal, not observations that the extra relationship is necessary. The competing counts make different claims about what can be combined. One proposes a single rule that must predict both rebuilding a message and choosing between messages with the same independently calibrated settings. Another keeps rebuilding and choosing separate but attributes apparent extra families to condition-dependent errors in how meaning is scored. A three-family alternative adds a dependence of rebuilding on the surrounding mix of competing versions. A six-family alternative separates taking in a source, later accessing it, combining sources, tailoring an output to an audience, selecting a version and allocating future exposure; it proposes that their combination already explains overlapping-update effects. The four-family proposal therefore needs more than a difference between overlapping and sequential updating: its extra relationship must remain necessary against these alternative accounts on experiments not used to adjust them. The supplied endpoint gives a specific comparison with established partner learning and message history, but does not supply a complete test program that resolves every rival partition. The screened literature supports narrower ingredients. S1, a 2015 Cerebral Cortex study available here through one supplied full-text window, reports that brain activity associated with reasoning about other people's beliefs was recruited when conversational expectations were violated, with little evidence of partner-specific anticipation in those areas. That window bears on responding to a partner's naming behavior, not on learning separate partner and population meanings or transmitting conflicts created by overlapping changes. S2, the 1996 Journal of Experimental Psychology: Learning, Memory, and Cognition abstract, supports a role for conversational history and partner-specific agreements about descriptions in word choice; it does not establish transfer to unfamiliar partners or conflict from overlapping public updates. S3, the 2009 Topics in Cognitive Science abstract, reports analyses showing that adaptation to a particular partner need not occur late during understanding; it does not establish the proposed separation between partner and population learning or the need for four families. S4, the 2020 Frontiers in Psychology study, reports that four-year-olds tracked two partners' knowledge and used it while interpreting hesitant speech. That result supports sensitivity to individual partners, not the proposed population-learning structure or the inheritance of inconsistent shared rules. S5, the 2007 Animal Cognition abstract, describes observational evidence of orangutan leaf-carrying and suggests that seeing associates collect materials prompted copying; it does not test human shared interpretation rules or establish why overlapping changes would create a separate causal family. S8, the 2021 Journal of Comparative Psychology abstract, reports enhanced visual discrimination learning in chimpanzees hearing contextually appropriate food calls; it does not test rule conflict, accurate source memory or overlapping public updates. These two studies supply background about social learning, not evidence for the proposed four-family count.

   Stated in the chain

What is carried, and what is not. Four screened sources speak to one ingredient, people's sensitivity to particular conversational partners, and two provide background about social learning in other apes; none of those sources establishes the full proposed separation between learning about a partner and learning about a population. No supplied source establishes the sequence from overlapping changes to an inconsistent accepted version to its onward use despite accurate recall, or the claimed minimum of four necessary causal families.

What would make this wrong. The proposed extra shared-update family would be unnecessary if an independently fitted combination of established partner learning, effects of preceding messages, rule learning and exposure-allocation history predicted the supposedly distinctive excess conflict on the held-out tests. It would also fail its stated prediction if verified removal of incompatible overlap left the excess unchanged, or restoring that overlap failed to restore it, with accurate individual recall and the specified matching maintained. A four-family minimum would fail separately if partner-history effects could be merged into reconstruction without losing the relevant predictions. These outcomes would reject the endpoint's proposed causal division, not the existence of cultural transmission.

What it would change. If the predicted excess survives the matched controls and disappears and returns with removal and restoration of incompatible overlap, cultural transmission research would need to track how changes become jointly accepted, alongside what individuals remember and choose to repeat. Claims about persistence and change would then have to distinguish faithfully remembering each contribution from maintaining a compatible shared system. A four-family count would still require the other proposed relationships to remain independently necessary in the tested set of experiments. Even that result in an online game with invented labels and rules would not establish the same division for internet narratives, long-lived cultural practices, recommendation systems or communities outside the tested setting.