The Platform as Bundler Attribution Rate, Effective Distinguishability and the Market for Confidence: In What Sense a Short-Video Platform Is a "Success Doctrine"

A success-doctrine course can be read as a bundle of belief goods with three components: comfort, certainty and testable guidance. Beyond the mix of components, this paper introduces an independent dimension, attribution rate r_A: the probability that, once a failure occurs, it is attributed to the guidance component. When r_A is low, failures are absorbed by the certainty narrative ("you have not grasped the Way", "your level is not high enough"), and the system loses its ability to correct itself. In the series' notation, distinguishability I_d denotes the per-step Kullback–Leibler divergence between the regimes before and after a change; we show that the attribution rate acts as a multiplier, reducing the effective distinguishability available to the buyer to about r_A·I_d. Six results follow. First (tier B), r_A and the guidance component g are independent dimensions; their empirical positive correlation comes from seller incentives. Second (tier A), r_A governs the speed of learning: for a buyer who knows that failures are absorbed, the number of observations needed to tell a good method from a bad one scales roughly as 1/r_A; for a naive buyer who updates at face value, below a critical value (about 0.65 in the demonstration) practice deepens belief in an ineffective method. Third (tier B), the seller institution holds an internal denominator, but low r_A rewrites failures as "not yet completed" at the recording stage and contaminates the denominator itself. Fourth (tier A, within a stated parameter region), when low r_A coexists with expensive tiered courses, a failed member can be worth more in expected revenue than a successful one, and the derivative of revenue with respect to the true success rate is negative. Fifth (tier A, minimal model), distribution that optimises immediate engagement makes the share of long-feedback-lag content decay geometrically. Sixth (tier A within a stated model, tier B for platforms), as long as immediate engagement rises as r_A falls, competition among creators drives r_A to its lower bound, and even knowledge content that once had high r_A is forced into the grammar of success doctrine. Hence a short-video platform is neutral in content but not in grammar: it is a machine for lowering the attribution rate, and with it effective distinguishability. It holds a vast denominator of attention, yet by design does not collect the denominator of users' long-run outcomes; to users it distributes low-r_A narratives, and to creators it is itself a bundle driven by survivor narratives. When many users receive narratives from a common source, the correlation ρ of their belief errors rises and, together with low r_A, weakens collective error correction. Once AI takes on both production and distribution, r_A may be lowered separately for each person, or raised to an unprecedented level, depending on which denominator is written into the objective. The philosophical corollary is that the greatest failure of information markets is that failure has nowhere to be booked, so effective distinguishability collapses; lies are only the easier case. Six testable hypotheses requiring no human subjects are registered, and remedies are proposed centred on raising r_A, disclosing cohort outcome denominators, and constraining tiered selling and platform objectives.