1: \begin{definition}[Preliminaries]

2: Let $(\Omega,\s{F},\P)$ be a probability space.

3: Let $\c{M}_s(\N)$ be the space of signed measures, equipped with the topology of weak convergence.

4: Let $\c{M}_1(\N) \subset \c{M}_s(\N)$ be the space of probability measures over $\N$, and identify $\c{M}_1(\N)$ with the probability simplex by the homeomorphism $\c{M}_1(\N) \isom \{\v{x} \in \ell^1 : \forall i, x_i > 0, \sum_{i=1}^\infty x_i = 1\}$.

5: Let $N\in\N$, let $\v{x} \in \N^N$, and let $\v{l} \in N^N$ be its empirical counts, defined by $\v{l} = \sum_{i=1}^N \v{1}_{x_i}$ where $\v{1}_{x_i}$ is equal to $1$ for coordinate $x_i$ and $0$ for all other coordinate.

6: Let $\gamma \in \R^+$.

7: Recall that $\N^N$ and $\c{M}_1(\N)$, endowed with the discrete topology and topology of weak convergence, respectively, are both Polish spaces---hence, the Disintegration Theorem (\textcite{ambrosio07b}, Theorem 5.3.1; \textcite{bogachev07b}, Corollary 10.4.15) holds in both spaces.

8: We associate each random variable $y:\Omega\-> Y$ with its pushforward probability measure $\pi_y(A_y) = \sbr{y_*\P}(A_y) = \P[y^{-1}(A_y)]$, and each conditional random variables $\theta\given y:\Omega\x Y \-> \Theta$ with its pushforward regular conditional probability measure $\pi_{y\given\theta}(A_y \given \theta) = \sbr{(y\given\theta)_*\P}(A_y) = \P[(y\given\theta)^{-1}(A_y)]$, where the preimage is taken with respect to $y$.

9: \end{definition}

10: