By destroyed I mean the data directory is wiped as a first step towards a full state transfer.

By the time we restored stable shell access to the nodes, after a lot of intermittent networking, all 3 nodes had indeed wiped themselves. The cluster was no more.

There may have been of course also something to do with SeveralNines CC that we were trialling at the time. Perhaps it was trying to do something hasty as you put it.

In any case, a lot of moving parts to understand fully.