The transcoder's output carries only the full mix, so a WebRTC viewer who picked a quality on the node that transcoded it got no stems, and silence after mixing one in. Packetize those renditions with the source's Opus tracks, the way replicated renditions already are; the pairing is now shared by both.