The rebuild that stalled at 62%.

The server arrived Monday morning — sent up from Paisley overnight, bays photographed and disks labelled by a commendably careful IT contractor. History: disk 2 had failed Tuesday; the spare kicked in; the rebuild crawled; then disk 4 dropped out at 62% on Friday and the volume went with it. Nobody had forced anything since — which, in RAID recovery, is the difference between a solvable puzzle and a scrambled one.

Nothing exotic: the disks were same model, same batch, same power-on hours — and a rebuild is the hardest week of a disk’s life, reading every sector of every survivor flat-out. Disk 4 had been quietly nursing pending sectors for months; the rebuild’s sustained load finished it. The array’s real state was now split across time: three disks consistent to Friday, one frozen at Tuesday, and a half-written spare muddying the middle.

All five disks — including the “dead” pair — were imaged individually first, the weakest on the gentlest settings. Disk 4 gave up nearly everything once read on hardware that doesn’t panic at slow sectors. Then the detective work: parameters confirmed from the metadata, the spare’s partial rebuild excluded, and the array solved virtually from the Friday-consistent set — parity filling disk 4’s few genuine gaps. The physical server was never asked to rebuild again.

The reconstructed volume mounted cleanly; shares, databases and a decade of job files verified against the directory tree. Everything shipped back on encrypted media Sunday evening with a written account of which disks to retire (all of them, frankly — same batch, same age). Multi-disk band, quoted to the job at diagnosis, priority track because a business was down. The firm’s Monday started on time.