The physical media
Platters and heads, or NAND flash. Where this is damaged, the affected regions are gone — nothing recovers a surface that no longer holds a magnetic pattern.
Hardware
Nobody knows how many until it stops, so the whole discipline is organised around spending them deliberately — one complete read-only copy first, then every repair and experiment against that image rather than your device.
A deteriorating drive may have exactly one complete read left in it. Spending that on a full image means every attempt afterwards happens on a copy, costs nothing, and can be repeated as often as needed.
Almost everything else follows from this, and it is the difference between a laboratory and a piece of software.
A failing device may have a limited number of good reads left in it. Possibly hundreds, possibly one. Nobody knows which until it stops, so the entire discipline is organised around spending those reads deliberately rather than discovering how many there were.
Which means the first real step is always the same: make a complete copy, read-only, before attempting anything else. Every reconstruction, every repair, every failed experiment then happens against that image. The original is never written to and never has to survive a second attempt.
The practical consequence matters more than it sounds. On an image, a failed attempt costs nothing. Running software on the original drive has no such property — and that, rather than any particular technique, is the substantive difference.
In that order, and the first one decides the other three.
Mechanical, electronic, firmware or logical. That single decision determines the route, the equipment, the price band and the realistic odds — so nothing sensible happens before it. It is why the diagnostic comes first and why it is free.
Donor head assemblies, board repair with the original ROM adaptives migrated across, service-area rebuilds, controller access in vendor mode. Enough to make the device readable once. Not enough to make it usable again, which is not the goal.
Cloned on hardware built for failing media: per-head targeting, timeouts in milliseconds rather than seconds, unreadable sectors logged and skipped rather than retried indefinitely. A fast first sweep banks everything that reads immediately; later passes revisit only the gaps.
File systems repaired from their own backup structures, arrays reassembled with the geometry derived from the data, files carved by signature where no structure survives at all.
Recovered files opened in the applications that made them. Anything unreadable is reported by name rather than returned as something that opens to nothing.
Understanding these explains why two drives with identical symptoms can need completely different work.
Platters and heads, or NAND flash. Where this is damaged, the affected regions are gone — nothing recovers a surface that no longer holds a magnetic pattern.
Drives run software from a reserved service area; SSDs keep translation tables. When those become unreadable the device cannot describe itself, while your data sits entirely untouched behind it.
Stripe geometry, member order, parity rotation, vendor pools. Reconstructed by analysing the disks rather than trusting the controller that already failed.
The map from your folders to physical locations. Damaged here, everything is present and unreachable — and most file systems keep backup copies of their critical structures.
Because a page explaining how recovery works should also explain where it stops.
Overwritten data. Once something has been written over your file, that portion is gone. There is no residual magnetic trace to read back on any drive made in the last twenty years, whatever older articles suggest.
Encryption without the key. BitLocker, FileVault, hardware-encrypted drives. Where the key is lost we can attempt password recovery, which works on weak or partly-remembered passwords and not against a long random one.
TRIMmed SSD deletions. The controller erases those blocks physically, within minutes, on a powered drive. That erasure is genuine.
Severely scored platters. Where a head has removed the magnetic coating, that band holds nothing. Undamaged areas of the same platter usually image normally, so the result is a partial recovery with the missing files named individually.
Everything else is a question of time and cost rather than possibility. Single drives in Manchester are from £300 +VAT after a free 48-hour diagnostic.
The fault is classified first, then the device is repaired only as far as reading requires, imaged read-only, and rebuilt from that image. The principle underneath is that a failing device may have few good reads left, so they are spent deliberately on one complete copy.
Because a failed attempt on an image costs nothing, while a failed attempt on the original may cost you the drive. Every reconstruction runs against the copy, so nothing anyone tries can reduce what is recoverable.
Repair aims for a working device; recovery aims for one readable pass. A drive fitted with donor heads is a recovery vehicle rather than a disk to trust — it is read once and retired.
No. Once something is written over your file, that portion is gone. There is no residual magnetic trace to read back on any drive made in the last twenty years, despite what older articles suggest.
With the password or recovery key, yes — encryption is no obstacle at all. Without it, healthy media yields nothing readable. We can attempt password recovery, which succeeds on weak or partly-remembered passwords and not on long random ones.
The free diagnostic is normally back within 48 hours. A single drive typically completes in three to four working days from approval; donor parts, heavy imaging and array work take longer, and you get the timescale with the quote.