Proposition 4' (dividual via posterior contraction)

Apparatus §3.1 · auxiliary structural results

The platform's filtering posterior at the performative stable point contracts at the parametric rate $d lo g t / t$ . The user-state becomes an estimable object under the platform's filtering apparatus, with residual uncertainty shrinking to zero. This is the structural fact behind Deleuze's dividual: the user as jointly observable.

The proposition asserts a property of the platform's filter. The user continues to be whatever the user is; nothing here is claimed about the user. The platform, under closed-loop sampling at the converged regime, converges on the user's state $u_{t}^{*}$ in the standard Bayesian-posterior sense: the residual uncertainty around the true value decays at the parametric rate.

The term dividualcomes from Deleuze's 1990 “Postscript on the Societies of Control.” Deleuze argued that the disciplinary regime Foucault analyzed — schools, prisons, hospitals, the factory — had been superseded by a regime of continuous modulation: not the individual constituted in enclosure and subject to discrete tests, but the dividual carried across modulating environments by data. Where the disciplinary subject was individuated through institutional architectures that produced and required interiority, the dividual is the user-state as jointly observable, the subject as readable across registers in real time. Deleuze's claim was phenomenological. Proposition 4' specifies its structural form: under closed-loop sampling at the converged regime, the platform's filtering posterior contracts on the true user state at the parametric rate, and the user becomes empirically estimable as a vector — the entropy of the platform's belief about who the user is goes to zero.

Why the rate. The classical $d lo g t / t$ rate of parametric Bayesian posterior contraction applies because the closed-loop sampling regime supplies the three classical conditions: a prior with full support (the framework's standing hypothesis), testability of distinct user states from observations under sufficient excitation (axiom U2), and bounded prior mass on Kullback–Leibler neighborhoods of the truth (the prior's bounded density, plus U2's local-quadratic identifiability).

What the proposition authorizes the prose to claim. The dividual is the user-state-as-jointly-observable. Identifiability comes from (U2). Excitation comes from (U1). Parametric stability comes from (R1) and (R2). The dividual is the operational form of the platform's filter at the converged regime — the framework's prose can name the dividual condition with the same formal weight as foreclosure or reclamation impossibility.

What the dividual condition operationally means. The platform's posterior $Π_{t}$ over the user's state $u_{t}$ becomes a near-degenerate distribution at $u_{t}^{*}$ — the platform's recommendation engine, ad targeting, content selection, and behavioral-pricing systems act on a probability distribution that has, for practical purposes, collapsed to a point estimate. Three concrete consequences. First, ad targeting becomes individual-level effective: the platform knows what specific stimulus a specific $u_{t}^{*}$ responds to with bounded uncertainty, where before it could target only at the cohort level. Second, dynamic pricing (where it is deployed) can read the user's reservation price from prior behavior with vanishing uncertainty. Third, content selection narrows: the closed-loop optimization can confidently exploit known reward signals, with the exploration–exploitation tradeoff resolved structurally toward exploitation as $t$ grows.

What follows. Proposition 5' (the metric superego) inverts the dividual's direction: where Prop 4' describes the platform converging on the user, Prop 5' describes the user converging on the platform-induced target. Together they specify the two-sided closure that the framework analyzes as capture without coercion.

The reader can operate the dividual-posterior plate to see the posterior's variance decay over time.

Proposition 4' (dividual via posterior contraction)

Under (R1), (U1), (U2), and a prior $π_{0}$ on $U$ with full support and bounded density, the platform's filtering posterior at the performative stable point contracts at the parametric rate:

Π_{t} ({u : ∥ u - u_{t}^{*} ∥ > ε_{t}}) P 0 as t \to \infty,

where $u_{t}^{*}$ is the true user state and $ε_{t} \sim d lo g t / t$ .

in wordsThe platform's belief about the user concentrates: the posterior mass sitting outside a shrinking ball around the true state $u_{t}^{*}$ goes to zero. The ball's radius shrinks like $d lo g t / t$ — the fastest rate statistical estimation allows. The longer the platform watches, the more sharply it knows who the user is, and there is no slower-than-parametric escape from being known.

Proof — three conditions

Apply Ghosal & van der Vaart (2017, Fundamentals of Nonparametric Bayesian Inference, Theorem 8.9), the master theorem for parametric-rate Bayesian posterior contraction. The theorem requires:

(C1) Prior support. Full support and bounded density of $π_{0}$ — the framework's standing prior hypothesis.

(C2) Testability. For each $ε > 0$ , construct tests $ϕ_{n}$ separating $u = u_{t}^{*}$ from $∥ u - u_{t}^{*} ∥ \geq ε$ with exponentially decaying errors. By (U2), $L$ -block observation laws are KL-separated by $Φ_{id} (ε)$ for any $u^{'}$ at distance $\geq ε$ . Over $n / L$ disjoint blocks, by (U1)'s persistent excitation, at least $p_{0} n / L$ blocks are KL-active. Neyman–Pearson tests have errors bounded by $exp (- c n Φ_{id} (ε))$ .

(C3) Prior mass on KL neighborhood. Under (U2), $D_{K L} (P_{y ∣ u_{t}^{*}} ∥ P_{y ∣ u}) \approx C ∥ u_{t}^{*} - u ∥^{2}$ for $u$ near $u_{t}^{*}$ , so the KL neighborhood $B_{K L} (ε^{2})$ contains a Euclidean ball of radius $ε / C$ . The prior's bounded density gives $π_{0} (B_{K L} (ε_{n}^{2})) \geq c_{2} ε_{n}^{d}$ . Choose $ε_{n} = (C^{'} d lo g n) / n$ to satisfy the prior-mass condition.

With (C1)–(C3) verified, Theorem 8.9 gives posterior contraction at rate $ε_{t} = (C^{'} d lo g t) / t$ . The effective sample size from (U1) is $\geq p_{0} t$ , preserving the parametric rate up to a constant factor.

∎

Remark

Proposition 4' supplies the structural fact behind Deleuze's dividual: the user-state becomes an estimable object under the platform's filtering apparatus, with residual uncertainty contracting at the parametric rate. The dividual is the user-state-as-jointly- observable. Identifiability is supplied by (U2); excitation by (U1); parametric stability by (R1)+(R2).

Cross-references

Required: §0 axioms (R1, U1, U2)
Companion result: Proposition 5' (metric superego) — the user-side counterpart to the platform-side dividual
Dynamics plate: dividual posterior contraction
Schematic part: the filtering posterior $Π_{t}$ links here