I think you refer to this forum thread for the 10% boundary. The 10% is chosen mainly from the perspective of node rewards, and how much average downtime during the reward period - e.g. a month - would be reasonable. It is currently (not yet) related to the actual replacement of nodes.
For replacement of nodes, DRE normally runs the so called dre-heal check every Friday, to replace degraded nodes, and these nodes would then be swapped on Monday. So worst case if a node “would die” on Saturday, if would be replaced by the Monday in the week following, so be 10 days offline. On a monthly bases this would be something approximately10/30 is 33% unavailability that is technically allowed, but for the part above the 10% there would be a reward reduction (if that eventually gets implemented).
It would be good to have some criteria for when to replace unhealthy nodes. We now have:
- running the dre-heal every week on Friday, resulting in dead nodes being swapped on Monday
- if the situation requests, e.g. when there are 3 or more unhealthy nodes in a subnet, an urgent NNS proposal is submitted.
But we definitely could add some other criteria.