Celestron mounts have a known issue that the slew does not land exactly where intended. The sync is fine and the mount is capable of very accurate centering with plate solving and no synching, but I guess it is not yet implemented in SGP.
The current beta instructions describe synchless centering as something needed for mounts that don't like to be synch'd - but there are also mounts like the cgem that can synch just fine, but the slew does not happen exactly as sgp wants it to, and that limits the centering accuracy when using the sync/slew approach.
The problem is with the slew - not the sync.
So - for cgem and other celestron mounts - sync is fine and the mount's model is fine and the goto accuracy is good - but the sgp centering routine will not converge and will fail in the manner you describe - until syncless centering is implemented for it.
Frank