I'm studying the specification of Kyber that was submitted to NIST PQC Round 3. However, I cannot figure out how they compute the error probability $\delta$ for Kyber 512, 768 and 1024. I have read the Kyber paper (written with respect to round 1), especially its Theorem 1 (correctness theorem). But the paper did not explain how to compute $\delta$; instead, the author just stated that they "provide with [the] software" a Python script to compute it (which was in neither the paper's appendix nor, apparently, the software's Git repo).
$$\begin{array}{c} \text{Table 1: parameter sets for Kyber} \\ \hline \begin{array}{l c} & n & k & q & \eta_1 & \eta_2 & (d_u,d_v) & \delta \\ \hline \text{Kyber512} & 256 & 2 & 3329 & 3 & 2 & (10,4) & 2^{-139} \\ \text{Kyber768} & 256 & 3 & 3329 & 2 & 2 & (10,4) & 2^{-164} \\ \text{Kyber1024} & 256 & 4 & 3329 & 2 & 2 & (11,55) & 2^{-174} \end{array} \\ \hline \end{array}$$
Can one explain how $\delta$ is computed? How accurate or "tight" are these $\delta$ values, and how would one evaluate their correctness?
