A key principle in string processing is local consistency: using short contexts to handle matching fragments of a string consistently. String synchronizing sets [Kempa, Kociumaka; STOC 2019] are an influential instantiation of this principle. A $τ$-synchronizing set of a length-$n$ string is a set of $O(n/τ)$ positions, chosen via their length-$2τ$ contexts, such that (outside highly periodic regions) at least one position in every length-$τ$ window is selected. Among their applications are faster algorithms for data compression, text indexing, and string similarity in the word RAM model. We show how to preprocess any string $T \in [0..σ)^n$ in $O(n\logσ/\log n)$ time so that, for any $τ\in[1..n]$, a $τ$-synchronizing set of $T$ can be constructed in $O((n\logτ)/(τ\log n))$ time. Both bounds are optimal in the word RAM model with word size $w=Θ(\log n)$. Previously, the construction time was $O(n/τ)$, either after an $O(n)$-time preprocessing [Kociumaka, Radoszewski, Rytter, Waleń; SICOMP 2024], or without preprocessing if $τ<0.2\log_σn$ [Kempa, Kociumaka; STOC 2019]. A simple version of our method outputs the set as a sorted list in $O(n/τ)$ time, or as a bitmask in $O(n/\log n)$ time. Our optimal construction produces a compact fully indexable dictionary, supporting select queries in $O(1)$ time and rank queries in $O(\log(\tfrac{\logτ}{\log\log n}))$ time, matching unconditional cell-probe lower bounds for $τ\le n^{1-Ω(1)}$. We achieve this via a new framework for processing sparse integer sequences in a custom variable-length encoding. For rank and select queries, we augment the optimal variant of van Emde Boas trees [Pătraşcu, Thorup; STOC 2006] with a deterministic linear-time construction. The above query-time guarantees hold after preprocessing time proportional to the encoding size (in words).