The case
A20 §2: “Imagine a regulator weighing two competing accounts of an industry’s risks. At first, both have comparable standing. One industry’s representatives gain an initial informational advantage: perhaps they have better access to the regulator, provide more relevant technical material, or frame the problem in terms the regulator already favors. That interpretive advantage increases their credibility. Credibility then yields greater influence over which evidence the regulator considers and which experts it trusts. If subsequent credibility is assessed through the increasingly dominant interpretation, an endogenous amplification loop becomes possible.”
Held fixed: the regulator’s formal procedure, its authority to decide, and its inferential competence. It updates correctly on what it reads. Varied: who supplies the evidence and how the evidence is read.
In CASE-099’s terms, m is the lean of the regulator’s reading toward one account, and a the standing of the party that account favours. β is that standing shaping what is read (whose material, whose experts). γ is the reading shaping whom the regulator credits. The symmetric state is unstable when βγ > (1 − α)(1 − δ) (CLAIM-330). If the regulator scores the parties against evidence that the favoured party supplies, so that the prevailing reading bends the evidence more than it bends the parties’ predictions, κ′ > κ̄, and γ > 0 (CLAIM-340). If it scores them against evidence neither controls, γ is near zero or negative.
Variants.
- A one-time perturbation (A20 §3, reflexive control). An outside actor shifts the regulator’s reading once, m₀ → m₀ + ε. “If the target’s interpretive and authority dynamics are stable, that perturbation may disappear. But if the interpretation–authority loop is unstable, the perturbation can grow. The influencing actor need not continuously control the regulator’s decisions.”
- A press as the evaluator (A21 §3). The evaluator is a newsroom whose conditions of production the evaluated party’s decisions affect, and the loop runs between two institutions: “Public interpretation → Political and regulatory decisions → Distribution of resources and authority → Conditions of future media production → Public interpretation.” A21: “the potentially captured institution is the institution that should help evaluate another institution’s exercise of authority.”
- An objective that gives one property no weight (A26 §4). The regulator has all the evidence, and decides by U_λ(a) = v₁(a) + λv₂(a) with λ = 0. “Even perfect information about both properties does not ensure the regulator will consider the second.” No evidence is selected; the criteria are.
What it can show
That each route the claim it illustrates names is coherent: influence on an evaluator’s decisions with no decision issued to it, through the evidence it reads, through how that evidence is read, or through the weights of its objective, while its procedure stays formally its own. A20 §3 ends: “An intervention from outside the organization can alter its internal allocation of practical authority, after which the organization reproduces the altered allocation through processes that appear to be its own.” The checks belong to CASE-099 and CASE-106. The two constructions of the third variant are elementary and were checked: with λ = 0 the argmax does not depend on v₂, and with U fixed, a* moves with the evidence.
What it cannot show
- That any regulator, newsroom or industry behaves this way. It is stipulated. A21’s caption: “A possible structural mechanism, not a claim that any particular outlet or institution follows these dynamics.” The documented episodes the replies cited were read on 2026-10-21, except Stigler’s (LIT-1324, Unavailable). The press variant now has two documented instances and one model, none with this case’s loop. Di Tella and Franceschelli’s Argentine newspapers give an association within newspaper and presidency, and A21’s caution, “evidence of an association, not by itself proof of a particular covert intent”, still applies (THEORY-851). Szeidl and Szucs’s Hungarian media give two-way favors at five events, with private advertisers as control (THEORY-850); when the 2015 fallout cut the reward, content favors stopped after a lag, a dependence that did not sustain itself once the resource link was cut, which in that instance counts against the amplifying reading. Besley and Prat’s model is a one-shot bargain, so it has the press variant’s dependence and not its loop (THEORY-837). Schram and Ule’s regulators are a laboratory study, not a documented episode (THEORY-856). No variant of this case with a documented loop has been read.
- That the loop between institutions is unstable under any stated condition. CLAIM-330 is a 2×2 linearization of one organization, and the two-institution loop has not been derived.
- That the favoured party’s account is wrong. A20 §9: “such a feedback loop may also reflect genuine learning, merited expertise, and beneficial specialization. Only independent standards and causal evidence can distinguish those from inappropriate capture.”