Why the people on stage talk over WebRTC through an SFU, why the audience watches a low-latency HLS stream through a CDN, and the trade-offs between the two.
Why our video server is created before each session and deleted after, what outlives it, and what an incident taught us about timeouts and scheduler design.
How a replayed recording behaves like a live session: one start time, a borrowed server clock, a player that puts itself back, and replayed chat timed to the video.