GPSDOs are often Frequency Lock Loops (FLL) i.e. counting, not Phase Lock Loops (PLL) i.e. comparing phase. That said, if the control system is optimized then the phase changes of even an FLL might be quite modest, slow and stable. I have a similar "small" GPSDO that includes an Oven Controlled Xtal Oscillator (OCXO), confirmed because I've looked inside. The OCXO is an old/used/surplus module installed on the circuit board that is otherwise populated with apparently new parts. A Temperature Compensated Xtal Oscillator (TCXO) is certainly a '2nd class citizen', as they inherently have phase jumps when they drop pulses. OCXO is the way to go, larger or small box.
I'm not 100% sure that the PPS to phase comparison is entirely fair. If you look at the Allan deviation of a GPS receiver's PPS output on a log-log plot, you'll find it's a straight decade-per-decade line down and to the right. That is, the longer your averaging period, the better GPS is. But at very short tau, GPS has a jitter on the order of a dozen nanoseconds or more. Which is exactly what you're seeing. This jitter has two sources. One is based on the fact that the GPS receiver is probably running within a microcontroller with an unsynchronized execution clock. That fact will almost certainly result in some level of quantization error. This is because the PPS signal, being an output from the controller, is going to be by necessity synchronized with the execution clock, but that has no connection at all with anything in the GPS signal. Good GPS receivers will be able to tell you the magnitude of the quantization error of the most recent PPS, so that the GPSDO can compensate for that, even if they won't necessarily be able change the timing of the PPS signal. But even if you take QE into account, at low tau GPS will have several nanoseconds worth of jitter in it anyway. GPS is very bad at low tau, but is very excellent at high tau. That contrasts sharply with a typical OCXO, which has excellent stability at low tau, but will drift all over the place without being disciplined. This best-of-both-worlds is the whole idea behind a GPSDO. It's entirely possible (I have no idea) that the reason that the larger GPSDO's PPS is maintaining phase with the 10 MHz output is that it's not the GPS PPS, but rather just the 10 MHz clock divided down. If that's true, then the fact that it's maintaining lockstep phase tells you nothing at all about the stability of the GPSDO output. FWIW, I sell GPSDOs on Tindie.
Your explanation is completely plausible. What I’m more interested in is which XO type is used. My guess is the larger one uses OCXO, while the smaller is TCXO. That would explain the drift seen from the touching. For short term stability, OCXO is going to be much better.
It can be shown in both ways, but usally its easier to se differense in frequency when you connect it in XY mode becasue the circle will show differnece easier by moving around and be steady if both frequencys are exactly the same, the speed of the circle move is easier to spot with eye it called lissajous curve.
второй маленький видимо использует numerical controlled oscillator для формирования опорной частоты, методом пропуска импульсов, по этому идет дрожание фазы в контуре фапч после умножения частоты.