Why it is a question
A15: “Does coupling the construction of meaning to the allocation of interpretive authority produce causal modes, attractors, or transitions that cannot be reproduced as economically by established models of opinion formation with evolving influence networks?” It is the explanatory-novelty half of A15’s question; the other half, whether such modes persist through turnover and individuate an organization, is QUESTION-115 and QUESTION-118. A15: “We can establish the first without establishing the second.”
The candidate predecessors, as A15 described them (registered, unread):
- Reflected appraisal (Jia, MirTabatabaei, Friedkin and Bullo, LIT-1146): opinions converge on each issue, and each member’s social power on the next issue is set by its influence on the last. A15’s mapping: the record’s momentary endorsement, trust and authority (CLAIM-277). The difference A15 claims: the record’s operators decide what is retrieved, what counts as evidence and how the next query is framed, and their authority is indexed by role (CLAIM-281).
- Adaptive networks (Holme and Newman, LIT-1128; Gross and Blasius, LIT-1139): opinions and ties co-evolve, with a phase transition between regimes.
- Persuasion bias (DeMarzo, Vayanos and Zwiebel, LIT-1134): agents who do not discount repeated information give social position influence apart from accuracy. A15: “Endorsement-mediated reinforcement could sometimes amount to repeated counting of correlated evidence.”
The record already holds two that A15 named as new: Golub and Jackson’s naïve learning (THEORY-321, LIT-999, read), on which A15 rested CLAIM-276’s fidelity, and Kirby, Cornish and Smith’s cumulative cultural evolution (LIT-771, read), on which it rested selection for transmissibility over accuracy. Its readings of Zollman (THEORY-143) and Holman and Bruner (THEORY-146) are right.
This is what CLAIM-278 demands of the operator framework: an outcome it forbids. If a reduced CASE-092 is a DeGroot–Friedkin system, A15: “it would let us adopt an established stability analysis and reserve the new work for the mechanisms that actually differ.”
What would count as an answer
A derivation placing a reduced CASE-092 inside one of these models, or a phenomenon from A15’s suite (capture without fidelity, hysteresis, role-sensitive recovery, persistence through operator turnover) that the coupled meaning–authority model produces and they cannot at matched complexity. CLAIM-330’s instability condition is the linear part any of them would have to reproduce.
Note of 2026-10-18: a crossed ablation, and the condition it must meet
A32 of the exchange of 2026-10-17 proposed a design for this question while applying the coupled model to learning: crossed ablations in a learner with specialized operators, an influence-allocation mechanism and an internal estimate of strain, in four conditions (strain represented or not, able to change endorsement or not). Its prediction is not lower loss in the fully coupled condition but “distinctive persistent regulatory modes: coordinated shifts in which operators are recruited, which evidence is inspected, and which adaptation strategies are maintained”. And it named this question’s condition without citing it: “the model must make a prediction that its constituent frameworks do not make separately.”
The design answers this question only if a persistent mode appears in the coupled condition that a model of evolving influence with fixed operators does not produce at matched complexity. The comparison classes are the ones above, and CLAIM-297’s defeater, reproduction by a single adaptive-retrieval process, is the same test.
Note of 2026-10-21: a fifth candidate predecessor
A25 §2 of the exchange of 2026-10-20 asked this question again without citing it: “whether our second authority variable changes the dynamics in ways that cannot already be represented as endogenous network weights, strategic signaling, or standard adaptive learning”, and “CLAIM-330’s instability condition is valid mathematics. Its novelty cannot rest on the eigenvalue calculation itself.” It adds a candidate: Mostagir, Ozdaglar and Siderius, “When Is Society Susceptible to Manipulation?” (Management Science 2022, doi:10.1287/mnsc.2021.4265), now registered unread as LIT-1318. A25 says it “concerns strategic interference with social learning, rather than a dynamically evolving allocation of authority over interpretation”. That is the third exchange to pose the question, after A15 of 2026-10-16 and A32 of 2026-10-17. The predecessors registered here (LIT-1146, LIT-1128, LIT-1139, LIT-1134) are still unread. A25’s other first-ranked work, Weatherall, O’Connor and Bruner, is registered as LIT-1341.
Both were sought on 2026-10-21. Mostagir, Ozdaglar and Siderius is Unavailable: copies exist (Deep Blue, CORE, an author’s site) but none could be fetched here, so the check against it stays open. Weatherall, O’Connor and Bruner was read in full (NOTE-1038): its dynamics are open-loop, with no coupling from the audience back to the evidence or the scientists, so it is not a predecessor of the second authority variable; A25’s “alongside” fits. The question stays Open.
Note of 2026-10-25: endogenous influence and repeated messages, again
From A36, A38 and A43 of the exchange of 2026-10-24, continued.
- Endogenous influence. A36 proposed an influence matrix that evolves, under which dL = ℓᵀdx + (dℓ)ᵀx, and a loop from access through influence, observed performance and endorsement back to access, as “a microscopic realization of the meaning–authority–legibility loop”. The nearest exact model is reflected appraisal (THEORY-916), in which ℓ is fed back as self-weight. Not cited.
- Repeated messages. A38 reported that a population counting forwarded
messages as new evidence was more confident and less accurate than one
that tracked sources: 98.4% against 95.6% confidence, 95.0% against 97.0%
accuracy, with about 16 repeated observations counted per episode, over
eight seeds. That is persuasion bias (LIT-1134), from a simulation the
record does not hold. A43 added that distinct source identities need not
be independent (its
correlation_group), which is also what changes the collective noise in CASE-160 from Σℓᵢ²σᵢ² to ℓᵀΣℓ.
The question is unchanged.
Note of 2026-10-26: phenomena shown, no comparison made
Of A15’s suite, the Perspective Population Lab now shows capture without fidelity and history dependence in stipulated models (CASE-162), and responses to intervention that ordinary trajectories conceal (CASE-161). None was set against DeGroot–Friedkin, an adaptive network or persuasion bias, so none bears on “as economically”.
A comparator pilot (dmarx/PerspectiveLab@dce302b experiments/comparator_pilot/) adds a constraint on any such comparison. A generic model with the same transition kernels makes the same predictions, by induction in time. So the comparison has to be of restrictions, compression or transport, with knowledge of the candidate mechanisms priced. The pilot’s report says it does not settle this question. The question is unchanged.
The lab’s v2.3 (dmarx/PerspectiveLab@e9efcf4 experiments/matched_comparator/) is the first comparison on equal terms of data and fitted parameters, though not yet against the models this question names. An 8-coefficient structural predictor and generic system identifiers are fitted on the same source transitions and tested on withheld production laws and a withheld dependency graph. A same-basis state-space model ties exactly. A generic 20-coefficient polynomial does better on the unseen bottleneck law with the mixed graph, MSE 0.001865 against 0.004514 over 24 seeds. An 8-coefficient generic polynomial leads in one familiar regime, within seed noise. The structural basis is not priced in description length, and DeGroot–Friedkin, adaptive networks and persuasion bias are still not among the rivals.
Note of 2026-10-29: three of the predecessors, read
Holme and Newman (LIT-1128), Gross and Blasius (LIT-1139) and DeMarzo, Vayanos and Zwiebel (LIT-1134) are now read, so the summary’s “registered unread” and the note of 2026-10-21’s “still unread” no longer hold for them. Each sets a different floor.
- Holme and Newman (NOTE-1385). It has no influence weights, no authority, no evidence and no turnover: opinions are labels, every tie is equal, and the dynamics is absorbing. What it reproduces economically is fragmentation into like-minded camps against one dominant camp, controlled by the ratio of selecting one’s contacts to being influenced by them, with a critical point between. It cannot express an outcome that depends on who has authority over whom. An answer here that rests on fragmentation alone would be matched by it.
- Gross and Blasius (NOTE-1394). The review supplies vocabulary and a list of mechanisms, not tools: no model in it has authority or role, and its only weighted, directed cases are coupled oscillators. Two items bear. Like-and-like reinforcement producing persistent leader and follower classes is a predecessor to beat if endorsement concentration is offered as new. Moment closure, a few variables for state plus topology, is the kind of economical reduction “as economically” invites. Its summary of Holme and Newman is wrong, so it should not be used as a pointer to that model.
- DeMarzo, Vayanos and Zwiebel (NOTE-1392). The fixed-weight model already gives one of A15’s suite, capture without fidelity: by its Theorem 2 an unbalanced network’s consensus departs from the precision-weighted optimum with no change in anyone’s accuracy. Its weights never move, so it has no hysteresis and no feedback from outcome to authority, and operator turnover would only change the listening matrix. Whatever the coupled model claims beyond position-weighted consensus has to come from moving weights.
Guida, Minutola and Presti, LIT-1218 (NOTE-1353), demand that an agent’s leverage be measured independently of how far its output spreads. That is the circularity guard an answer here needs.
The question is unchanged and stays Open.
Note of 2026-10-29: the like-and-like predecessor, read, and the moving-weights companion, not reached
- Ito and Kaneko, LIT-1842 (NOTE-1414), the like-and-like reinforcement the earlier note of this date takes from Gross and Blasius, is now read. It is closer to this question’s endorsement circuit than the review suggests. Each unit allocates a fixed budget of attention (its incoming weights sum to 1), moves attention toward those whose state resembles its own, and is driven by those it attends to. From identical units with no role primitive, it gives a persistent asymmetry of influence, a core of about 15 attended by most and attending almost only to itself, with links turning over, and an ablation shows that the coupling of state to allocation is needed. So an answer that offers persistent authority classes, or their need for the coupling of state and authority, as new is matched by this model. What it cannot express is what the question’s circuit adds: states with content that can be right or wrong, any evaluation of outcomes, and authority indexed by role or issue. The outcome to beat is concentration without regard to accuracy. That is also persuasion bias’s outcome (LIT-1134), and this model gets it with moving weights and hysteresis, which persuasion bias lacks.
- DeMarzo, Vayanos and Zwiebel 2001, LIT-1825, the companion to LIT-1134 in which whom an agent listens to is chosen by what listening earns, is registered and Unavailable: no copy of the working paper was found. As LIT-1134 describes it, it would be a predecessor with evolving influence that already gives capture without fidelity. Whether listening there is scored against the interpretation it made dominant, or by trading profits that depend on what the asset pays, can only be read from the paper.
The question is unchanged and stays Open.