yeah, you're right, it's sort of a brain in the jar problem, isn't it. but it does seem like it's not that hard to detect because of latency between A and B, though it would require a pretty accurate clock on GPS device.

No! Differential latency from the satellites is entirely how GPS works, and there is no way of having an accurate enough clock to be able to spot that without (a) it being an atomic clock and (b) synced to the satellite time at a known location.

But accuracy of GPS signal is in microseconds, so if you suddenly detect a change in level of milliseconds (latency between A and B) - that would totally imply spoofing shenanigans?

Yes, which would manifest as a sudden movement of place detected. What I'm not sure about is what this looks like as you approach the jammer and its signal gradually overwhelms some or all of the signals; does it just look like multipath error.

Obviously I don’t know what I’m talking about, but it seems like such attack is only undetectable by cold gps device that has been offline for longer than some period of time, long enough to explain time shift by clock drift.