A Synology pool is a stack. At the bottom, each drive carries three partitions: a small system partition mirrored across every drive, a swap partition likewise, and the large data partition. The data partitions are joined by mdadm into one or more RAID arrays, on SHR into several of different sizes so that mixed drives are used fully, and the arrays are joined by LVM into a volume that carries Btrfs or ext4. DSM assembles that stack at every boot, and Crashed is what it reports when a layer will not assemble: an array missing too many members, an LVM map it cannot read, a file system whose metadata fails its own checks.
The usual road to Crashed runs through Degraded. One drive fails, the pool is degraded and still readable, a replacement is fitted, and the repair reads every sector of every survivor to rebuild the missing one. On drives that have shared a case for four years, that is where the second failure is found, and DSM, which cannot rebuild from two missing members, marks the pool Crashed. The drive that dropped out first still holds every stripe written before it dropped; the drive that failed during the repair usually still reads on the bench with its bad areas taken last. Between them and the healthy members, the set is complete.
That is why the bench never asks DSM to try again. Every drive is imaged, the weak one slowly, and mdadm's own superblocks on the images give the array's order, chunk size, parity layout and the event count that says which member left when. The arrays are assembled in software, the LVM map read, and the Btrfs file system mounted read-only from the virtual volume, with its checksums used to say which copies of the metadata are good. Nothing writes to the originals, and nothing can be lost by trying.