Ty for the informal beta testing hahaha I've found that the connection drops from the server's heartbeat are treated as errors by my client. Pushing a fix in 0.0.6 to reconnect quick and reconcile missed msgs.
Another thing I'm noticing is the firehose endpoint is reconnecting every few seconds instead of being kept alive. Is this a limitation of some sort of the EventSource client you're using? Because it might omit some messages, the reconnection doesn't do any replay.