Ratchets, Rebates and Refunds Three Self-Reinforcing Loops in Success-Doctrine Organisations: the Instructor's Asymmetric Updating, the Accounting Identity of Multi-Level Membership, and the Armour of "Hard to Execute"

Beyond the three-component model of comfort, certainty and guidance, this paper studies three structural mechanisms of success-doctrine organisations. The instructor's asymmetric updating. Members measure their results against the instructor's promise; of the results below the promise, only a fraction a returns to the instructor's guidance ledger as “the method failed”, while the rest are rewritten by members, in the instructor's own words, as “I didn't do it right”. When feedback has a finite mean, an instructor who reads reports at face value has a belief that converges to the τ-expectile of the feedback distribution, τ = 1/(1 + a); the smaller a, the larger the bias; as a → 0 the bias is unbounded, and at a = 0 there is no finite steady state and belief only rises (Proposition 1, tier A). “The smaller a, the higher the belief” needs no moment condition: for the same sequence of feedback it holds at every step, heavy-tailed feedback included. The result does not require assuming that the instructor is insincere, but the model also does not distinguish deception from informed exploitation: an instructor who knows about the censoring and still makes promises on the censored feedback gives the same prediction. It holds only for the guidance component: the commission ledger is an exact record, and organisers know the distribution of referral income. If failure is judged against a fixed standard rather than against the promise, the bias is bounded and the ratchet disappears; if the promise follows belief only in proportion λ < 1, there is a finite steady state even at a = 0, and it is higher the closer λ is to 1. The asymmetric-updating ratchet comes from “the promise follows belief”. The thinner members' reserves, the more those who remain in the room seem to be improving (Proposition 2); once the inflow of new members stops, the room as a whole ages and the mood actually improves (§5.3). Multi-level membership and rebates. Referral rebates are bound by an accounting identity: at any moment, referral income covers on average across all members at most R times the mean membership fee, where R is the sum of the commission rates at all levels set by the commission rules; referral rewards paid from funds other than membership fees are counted separately (Proposition 3, tier A). The distribution of returns is decided jointly by joining position and recruiting ability; position is unobservable at the time of joining, and the joining position at which members break even in expectation also depends on the shape of the recruitment tree (Proposition 4); tiered membership turns an upgrade into the purchase of a commission rate, and an upgrade can pay for itself only if downlines upgrade too (Proposition 5). Rebates make every member with a downline a mini-instructor, and turn the early, upper nodes most likely to see that the method does not work into the people with the least economic freedom to say “it doesn't work” (Corollary 3.1). Refunds and failure. When the method is hard to execute in full, and failures without full execution can always be blamed on poor execution (a premise that is a behavioural assumption about the organisation's rhetoric), the attribution rate cannot exceed the fraction of failures in which full execution can be verified; once the execution standard is written in terms of mindset, the attribution rate can be set to zero after the fact (Proposition 6). “Everyone who followed it succeeded” is a pseudo-control selected on outcomes (Proposition 7). Failures must pass in turn through four filters — reattribution, upgrade monetisation, private settlement and silent exit — and these multiply with the reporting bias in the odds of the visible record (Proposition 8). Refunds conditional on full execution turn the right to exit into an obligation to keep executing (Proposition 9). When failures are told as cautionary examples, the corrective signal changes sign and the naive buyer's critical attribution rate rises (Proposition 10). The three mechanisms weaken, respectively, the vertical, lateral and outcome error-correction channels. When all three are weakened and the upgrade conversion rate and price ladder fall within the parameter region of [S1] Proposition 4, the derivative of organisational revenue with respect to member outcomes is negative: this is coupling of the opposite sign, not decoupling (Corollary 5.1); at this paper's demonstration prices it requires an upgrade conversion rate above 10%. The disappearance of oscillation cannot be read as stability; what can be read is recovery after a disturbance. The three loops correspond to the breaking gate, the formation rate and the attribution rate of the network model of Conviction Systems [S24], and the paper uses this to give the interface between the two papers (§5.6). The countermeasures centre on full-cohort intention-to-treat analysis with a comparison group, an unconditional cooling-off period and disclosure of the distribution of net rebates, and each measure is labelled with the parameter it acts on. The paper registers nine predictions and lists, for each, the strongest alternative explanation and a design to rule it out. Executions on external data: zero.