When originally implementing this I assumed seq was the reliable way to track cursor state, but it turns out it has a few issues. This change is prompted by the recent outage of wss://jetstream.us-east.bsky.network, it getting taken down completely, and then the domain name being repointed to west.
The recommendation from @calabro.io is to use timestamps over seq since seq can differ wildly between jetstreams, while time should be more closely matched. This also makes failover between instances less painful as you just need to wind back the clock a little bit, instead of switching between cursor types.
Thread for context https://northsky.app/profile/jola.dev/post/3mvwsrjvuhk2n