Fix \mathscr{S}, \nu, \tilde{\nu} and \{\mathscr{S}^\lambda\}, \mu as in the statement of the corollary. Conjugating the B^y_b by a unitary if necessary we assume without loss of generality that the reduced densities of |\psi\rangle on either subsystem are identical. For every \lambda define a symmetric strategy \tilde{\mathscr{S}}_\lambda = (|\psi_\lambda\rangle,A) and let \tilde{C}^\lambda be the associated correlation. We first show that
\ensuremath{\mathop{\textsc{E}}_{(x,y)\sim \tilde{\nu}}} \sum_{a,b} \big| C_{x,y,a,b} - \int_\lambda\tilde{C}^\lambda_{x,y,a,b}d\lambda\big | \,=\, \operatorname{poly}(\delta)\;.
For this we show that
\ensuremath{\mathop{\textsc{E}}_{x,y}} \sum_{a,b} \big| C_{x,y,a,b} - \mbox{\rm Tr}\big( (A^x_a)^T B^y_b(A^x_a)^T \rho \big)\big|\,=\, O\big(\sqrt{\delta}\big)\;,
and
\int_\lambda \ensuremath{\mathop{\textsc{E}}_{x,y}} \sum_{a,b} \big| \tilde{C}^\lambda_{x,y,a,b} - \mbox{\rm Tr}\big( (A^x_a)^T B^y_b (A^x_a)^T\rho_\lambda \big)\big|\,=\, O\big({\delta}^{c/4}\big)\;.
Together with (?), combining (?) and (?) through the triangle inequality gives (Proof). To show (?), write using the triangle inequality
\begin{aligned}
\ensuremath{\mathop{\textsc{E}}_{x,y}} \sum_{a,b} \big| C_{x,y,a,b} - \mbox{\rm Tr}\big( (A^x_a)^T B^y_b (A^x_a)^T\rho \big)\big|
&\leq \ensuremath{\mathop{\textsc{E}}_{x,y}} \sum_{a,b} \big| \langle\psi| \big( \ensuremath{\mathop{\rm Id}\nolimits} \otimes(A^x_a)^T- A^x_a\otimes \ensuremath{\mathop{\rm Id}\nolimits}\big) B^y_b (\ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^x_a)^T )|\psi\rangle \big| \\
&\quad + \ensuremath{\mathop{\textsc{E}}_{x,y}} \sum_{a,b} \langle\psi| \big(\ensuremath{\mathop{\rm Id}\nolimits}\otimes (A^x_a)^T \big)B^y_b \big( \ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^x_a)^T- A^x_a \otimes \ensuremath{\mathop{\rm Id}\nolimits}\big)|\psi\rangle\big|\;,
\end{aligned}
where we used the assumption that each B^y_b is a projection. Each of the two terms on the right-hand side is bounded in the same manner. We show how to bound the first:
\begin{aligned}
\ensuremath{\mathop{\textsc{E}}_{x,y}} \sum_{a,b} \big| \langle\psi| \big( \ensuremath{\mathop{\rm Id}\nolimits} \otimes(A^x_a)^T &- A^x_a\otimes \ensuremath{\mathop{\rm Id}\nolimits}\big) B^y_b (\ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^x_a)^T )|\psi\rangle \big|\\
&\leq \Big(\ensuremath{\mathop{\textsc{E}}_{x,y}} \sum_{a,b} \langle\psi| \big( \ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^x_a)^T - A^x_a\otimes \ensuremath{\mathop{\rm Id}\nolimits}\big) B^y_b \big( \ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^x_a)^T - A^x_a\otimes \ensuremath{\mathop{\rm Id}\nolimits}\big)|\psi\rangle \Big)^{1/2}\\
&\qquad\qquad\cdot \Big(\ensuremath{\mathop{\textsc{E}}_{x,y}} \sum_{a,b} \langle\psi| ( \ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^x_a)^T) B^y_b ( \ensuremath{\mathop{\rm Id}\nolimits} \otimes(A^x_a)^T )|\psi\rangle \Big)^{1/2} \\
&\leq \Big( \ensuremath{\mathop{\textsc{E}}_{x}} \sum_a \langle\psi|\big( \ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^x_a)^T- A^x_a\otimes \ensuremath{\mathop{\rm Id}\nolimits}\big)^2|\psi\rangle \Big)^{1/2} \cdot 1\\
&\leq \sqrt{2\delta}\;,
\end{aligned}
where the first inequality is Cauchy-Schwarz, the second uses \sum_a A^x_a = \sum_b (B^y_b)^2 = \ensuremath{\mathop{\rm Id}\nolimits}, and the last follows by expanding the square and using the definition of \delta. This bounds (?). Together with a similar bound for the term in (?) this shows (?). To show (?), we proceed similarly up until the last step (?), at which point a bound on
\int_\lambda \ensuremath{\mathop{\textsc{E}}_{x}} \sum_a \langle\psi_\lambda|\big( \ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^x_a)^T- A^x_a\otimes \ensuremath{\mathop{\rm Id}\nolimits}\big)^2|\psi_\lambda\rangle
is required. To obtain this, we first note that
\int_\lambda \ensuremath{\mathop{\textsc{E}}_{x}} \sum_a \langle\psi_\lambda|\big( \ensuremath{\mathop{\rm Id}\nolimits} \otimes (A^{\lambda,x}_a)^T- A^{\lambda,x}_a\otimes \ensuremath{\mathop{\rm Id}\nolimits}\big)^2|\psi_\lambda\rangle\,=\,0\;,
because A^{\lambda,x}_a is supported on the support of \rho_\lambda, which is totally mixed on its support. Moreover, forming the difference we have (using that \{A^{\lambda,x}_a\} and \{A^x_a\} are projective)
\begin{aligned}
|\eqref{eq:c2-2e1}-\eqref{eq:c2-2e2} |
&= 2 \int_\lambda \ensuremath{\mathop{\textsc{E}}_{x}} \sum_a \Big(\langle\psi_\lambda| (A^{\lambda,x}_a - A^{x}_a)\otimes (A^{\lambda,x}_a)^T|\psi_\lambda\rangle + \langle\psi_\lambda| A^{x}_a \otimes (A^{\lambda,x}_a- A^{x}_a)^T\otimes |\psi_\lambda\rangle \Big)\;.
\end{aligned}
Each of the two terms on the right-hand side is bounded by an application of the Cauchy-Schwarz inequality followed by (?). This shows (?), and hence (Proof).
Having established (Proof) we now prove
\int_{\lambda}\ensuremath{\mathop{\textsc{E}}_{(x,y)\sim \tilde{\nu}}} \sum_{a,b} \big| \tilde{C}^\lambda_{x,y,a,b} -{C}^\lambda_{x,y,a,b}\big | d\lambda \,=\, \operatorname{poly}(\delta)\;.
Combining (Proof) and (?) shows (3.3), concluding the proof. To show (?) we apply Lemma 2.10 for each \lambda to the strategy \mathscr{S} = (|\psi_\lambda\rangle,A^\lambda,B) here and \hat{A} in Lemma 2.10 is A here. Since |\psi_\lambda\rangle is maximally entangled and A^\lambda supported on its support, \delta in Lemma 2.10 equals 0. Applying the lemma followed by Jensen’s inequality gives
\begin{aligned}
\int_{\lambda}\ensuremath{\mathop{\textsc{E}}_{(x,y)\sim \tilde{\nu}}} \sum_{a,b} \big| \tilde{C}^\lambda_{x,y,a,b} -{C}^\lambda_{x,y,a,b}\big | d\lambda
&= O\Big(\Big( \int_\lambda \ensuremath{\mathop{\textsc{E}}_{x}} \sum_a \mbox{\rm Tr}\big( \big( A^{\lambda,x}_a - A^x_a\big)^2 \rho_\lambda \big)d_\lambda \Big)^{1/2}\Big)\\
&= \operatorname{poly}(\delta)\;,
\end{aligned}
by (?). This shows (?) and concludes the proof.