> Once you’re in the game it’s rock solid, which leads me to question the “client networking issue” explanation.
Could be different architectures between the "queue" and the in-game session.
Presumably with the in-game backends, the client gets told to connect to a specific backend (or through an LB with a sticky session), which keeps some of your active player state in-memory; and because of that, they'd want those backends to be robust to disconnection, allowing the client to reconnect and resume the existing backend session on the same server.
The queue, on the other hand, is the first-in-line protection for the game — equivalent to Cloudflare for an HTTP site. It has to deal with bots, DDoS attacks, etc. I would guess that they "solved" this by requiring a kind of proof-of-work — that "work" in this case being "keeping a socket open and responding to pings for the whole time you're in the queue."
If this were true, "client networking issues" would technically be the proximate cause of a client falling out of the queue and having to get back in line at the end. It'd be a problem the devs created for themselves, but it would indeed be the problem.