Following Dave's advice here: "Don't turn it on - take it apart!"
Meinberg Lantime M300 - Multi Reference NTP server. This server works as stratum-1 using a number of different input signals such as GPS, 1PPS, 10MHz, and IRIG/Timecode.
The loop time-constants seem to be very long, since the ADEV plot does not change much (at least not immediately) when connecting or disconnecting e.g. the 1PPS or 10MHz inputs. If I would be more patient I would measure it free-running and with all the different stratum-1 input signals for a few days each - that would maybe show some interesting things about how the internal OCXO is locked to the reference inputs. My assumption would be that IRIG would produce the worst ADEV, GPS second worst, 1PPS quite good, and the 10MHz input should produce an excellent ADEV with the shortest lock time-constant.