# Pastebin 8pgQvf8l
@jbenet I think that would correspond to a "overlay partition failure"  the discussion I am doing. The problem is that when you cluster the backups together, it is easy to isolate and eclipse them.

Using "random" locations like hash(id+"backup number x") is better for that purpose, but has a much higher maintenance overhead and a wide "vulnerable window" because it is not as easy to detect when a backup fails.