It looks like I never opened an issue about it, but I've been stung by Parthenon only warning on variables not found, too. IIRC this was partially because any read error (not just absence in the file) just generates the warning rather than a real error, which made it hard to debug restarting from face-centered fields, for example.
There are valid reasons to start with more Restart variables configured than are present in the restart file: in KHARMA for example, we'll evolve ideal MHD to some time, then switch to slower extended/viscous modeling once the run is far enough in that it matters. But, I think we should at least pare down the situations where this is allowed, e.g. only for restarts with a new input file, or only with some flag parthenon/job/restart_errors=warn set.
Originally posted by @bprather in https://github.com/parthenon-hpc-lab/parthenon/issues/1046#issuecomment-2051990245