A four-bay DS of mismatched disks, undone by its own helpfulness
A volume created using two 4TB disks and two 8TB disks had one dead member. An overnight rebuild process ran within DSM which finished in Volume Crashed. Underneath were two separate mdadm arrays. The first array used all four disks (it was still usable although degraded) and the second array used only the two 8TB disks. The arrays were reassembled sequentially outside of the device. After rolling the btrfs tree back one generation, every share came back.