但关键在于：这个提升等多仰仗强化学习的结果，而非来自蒸馏这个行为本身。
Shortcut Pre-calculation: For the most commonly used speed profiles, the travel costs (time/distance) for shortcuts between border points within each cluster are pre-calculated and stored. (Each border point effectively has an "entry" and "exit" aspect for directed travel).
，更多细节参见服务器推荐
GPT-5.2 Pro + Codex，这一点在快连下载-Letsvpn下载中也有详细论述
const text = await Stream.text(readable);
This step rapidly finds the optimal sequence of border points and shortcuts to get from your start cluster's periphery to your target cluster's periphery. It's incredibly fast because it's ignoring all the tiny roads within intermediate clusters.