From the code here, it appears that a syncing node uses the last applied time reported in heartbeats to determine whether the sync source is staler than the syncing node.

This data could be stale as heartbeats occur every two seconds.

The main concern here is if a syncing node is seamlessly replicating from a sync source, it wouldn't make sense to go off of stale heartbeat data and change the sync source while replication is ongoing.