Every Twitch VoD carries a recorded_at timestamp, the second that streamer hit Go Live. Two streamers in the same collab have two timestamps within seconds of each other. That is the sync signal. The sync editor reads those timestamps, anchors each VoD to its own wall-clock origin, and lines them up.
There is no offset slider to babysit, no audio fingerprint to fail on quiet scenes, no chat-spike correlation to misfire when one streamer has a quieter chat. The timestamps are the ground truth and they are accurate to the second. The alignment math runs entirely in your browser, no extra server pass.