[1] https://coverclock.blogspot.com/2017/05/my-stratum-0-atomic-...
[1] https://coverclock.blogspot.com/2017/05/my-stratum-0-atomic-...
Use the pulse output to lock a station clock for your local time needs (like syncing your data recording to wall time) and the sine output to lock your mixing oscillators.
A good GPS-disciplined OCXO can hold that accuracy over much longer intervals.
To take advantage of that accuracy you have to share the master reference oscillator between the telescope's clock tree and GPS receiver's clock tree. A commercial chipset's PPS output isn't going to cut it, but a dedicated amateur could definitely build a GPS-tagged radio telescope off of a common crystal like that.
The other thing that can help is to put receivers not too far apart. Say 25 dishes in an area with 100 meters diameter. That improves sensitivity by an order or magnitude, resolution by two orders of magnitude and the station can either be connected with a clock network, or (if using GPS) will at least see correlated clock errors that don't matter much in the correlation. The 1 part in 10^-7 accuracy of the GPS receivers is plenty accurate for the phase folding.
Conner-Winfield OH320 advertises an Allan Deviation of 2.7e-12 at one second. Its a very similar part, without all the hirel and space qualifications. 98 USD/ea at Digi-Key.
I agree wholeheartedly on the low-baseline setup case. An offline carrier phase position-time solution could stretch you out to much wider baselines, though.