I've met a strange situation. I have 3 stratum-1 ntp servers driven by holdover atomic clock, just mark them as A, B, C.
After several months, ntp server A has 2~3ms offset with the other two, and the clients(both ntpd and chrony) marked A as falseticker, which is expected behavior. Root dispersion of each ntp server is about 1.4~1.7ms.
But some of the chrony clients also mark B as falseticker, some marked B as combined, some marked B as selectable but not combined, and all chrony clients select C as current time source.
Offset between B and C are small enough(currently 200us~300us). According to selection algorithm, I think B shouldn't be falseticker? 'chronyc selectdata -a -v' shows like this:
On ntpd clients, the selection algorithm works as expected: A marked as falseticker, B and C are randomly selected and the other one marked as combined.