Who was the doofus who decided that the default layout for a Linux software RAID array puts the RAID superblock at the very end of the device or partition? The upshot of this is that the system can have a hard time telling the difference between an array that is built on raw devices, versus an array that is built on partitions that extend to the end of the devices. If you rely on the system's automatic array detection logic, it may get it wrong; this can have some rather unpleasant (and very puzzling) results.
I managed to get the MD subsystem on my work desktop into a very confused state because of this, and ended up accidentally corrupting (and needing to rebuild from scratch) the RAID-1 array I was setting up. No real harm done (no data lost), but man was that a head-scratcher!
When I finally figured out WTF was going on, I forced it to use "version 1.2" superblocks when I rebuilt the array. This puts the superblocks at a fixed offset from the *start* of the device or partition. Seems a lot more sensible to me!
Moral of the story: Always use the "--metadata=1.2" option if you're building a RAID array on partitioned devices!