Splinference shouldn't require "a kick in the pants" to start its embedding cycle if it loads after the store is populated. Upon startup, splinference should just look for keys that have this type applied (it's a cheap bitmask search) and start its epoch index from there to track changes.
On serverless, this happens via Nomic Atlas and fetch triggered by cron. Locally, seqlock protects against thundering herds here.