A mirror is not a copy
A sync client's job is to make every location look the same. Delete a folder on the laptop and it will do exactly that, everywhere, in seconds. One copy with several windows onto it is still one copy.
- 3CopiesOf the data. Distinct copies — not one copy with three windows onto it.
- 2MediaDifferent kinds. Flash, spinning disk, tape, object storage — failures they do not share.
- 1Off-siteAnother building. Not another room, not the shelf beside the machine.
The rule, and where yours stops short
How Many Copies Is Enough
Three copies of the data. On two different kinds of media. One of them somewhere else. It is a short rule, and almost every setup that fails was one line short of it. Build yours below and see which line is missing.
- 3 COPIES
- 2 MEDIA
- 1 OFF-SITE
- +1 OFFLINE
- +0 ERRORS ON RESTORE
The Backup Builder
A diagnosis, not a lecture. Tick what you actually have and the panel names the line you are short of.
Your inventoryTick everything you actually have
What you have
13 copies of the dataDistinct copies, not windows onto oneSHORT
12 different mediaFlash · disk · tape · object storageSHORT
01 off-siteAnother building, not another roomSHORT
0+1 offline or immutableOut of reach of a bad write or a bad actorSHORT
—0 errors on restoreYou have pulled real files back and opened themSHORT
Where this stops short
One copy. A sync folder and the machine it mirrors are the same data seen twice; a deletion on one is a deletion on both, within seconds.
The SMART Decoder
Pick an attribute. Plain language on what it means, whether it matters, and what to do about it.
Select an attribute
SMART attribute
Reallocated Sectors Count
What it means
Sectors the drive found bad and swapped for spares.
Does it matter
Yes — any non-zero value is a drive telling you it has started dying.
What to do
Back up now. A rising count means replace, not monitor.
Raw values are vendor-specific. Judge a drive on the direction a counter is moving, not on one reading.
SMART attribute
Current Pending Sector Count
What it means
Sectors that failed a read and are waiting to be reallocated.
Does it matter
Yes — arguably more than reallocated.
What to do
Image the drive before you do anything that writes to it.
Raw values are vendor-specific. Judge a drive on the direction a counter is moving, not on one reading.
SMART attribute
Reported Uncorrectable Errors
What it means
Reads the drive could not fix with its own error correction.
Does it matter
Yes.
What to do
Treat as data loss in progress. Restore from backup rather than repair.
Raw values are vendor-specific. Judge a drive on the direction a counter is moving, not on one reading.
SMART attribute
Spin Retry Count
What it means
Times the platters failed to reach speed on the first attempt.
Does it matter
Yes, on a mechanical drive — it points at the motor or bearings.
What to do
Power it down and image it on the next successful spin-up.
Raw values are vendor-specific. Judge a drive on the direction a counter is moving, not on one reading.
SMART attribute
Power-On Hours
What it means
How long the drive has been running.
Does it matter
On its own, no.
What to do
Use it as context for the counters that do matter.
Raw values are vendor-specific. Judge a drive on the direction a counter is moving, not on one reading.
SMART attribute
Wear Levelling Count / Percentage Used
What it means
How much of the SSD's rated write endurance has been consumed.
Does it matter
Yes, but predictably — this is a planned end, not a fault.
What to do
Plan the replacement. Expect read-only behaviour near the end.
Raw values are vendor-specific. Judge a drive on the direction a counter is moving, not on one reading.
SMART attribute
Temperature
What it means
Drive temperature.
Does it matter
Only at extremes, sustained.
What to do
Fix airflow. Do not panic at a warm drive under load.
Raw values are vendor-specific. Judge a drive on the direction a counter is moving, not on one reading.
The Failure-Mode Table
Five modes, and the honest first response to each. None of them begins with running a repair tool.
Mode
What you see
First response
Mechanical
Clicking, ticking, buzzing, failure to spin up.
Power down. Do not retry repeatedly. Image on the next good spin-up, or send it away.
Electronic
Completely dead, not detected, no spin at all.
Check power and cable first, then stop. Board-level work is a specialist job.
Flash wear
An SSD that has gone read-only, or throws uncorrectable errors.
Read everything off it now. Read-only is the drive's last cooperative state.
Controller / firmware
Detected but wrong capacity, or garbage where data should be.
Change nothing. Do not reformat to 'fix' it. Image the raw device.
Human
Deleted, overwritten, or synced away.
Stop using the volume immediately, then restore from a version rather than undelete.
STOP WRITING · FIRST AID
If a drive has just started misbehaving
- 01Stop writing to itEvery write reduces what can be recovered, including writes you did not order.
- 02Power it down if it is making noiseA head touching a platter destroys more with every rotation.
- 03Image the whole deviceWork from a copy so that a failed attempt costs you nothing.
- 04Work only from the imageRepair tools rewrite. Never point one at the original.
- 05Know when to stopIf the data matters more than the fee, further attempts are the expensive mistake.
The order is the safety rule. Every step that gets skipped removes an option you cannot get back.
Four things people count twice
RAID is uptime, not history
A mirrored NAS survives a disk. It does not survive a deletion, a bad write, a fire, or the controller that wrote garbage to both halves at once. Redundancy keeps a service running; backup keeps a past.
The desk is not off-site
A drive that lives beside the machine it backs up shares its power supply, its burglar and its flood. Off-site means a different building, and rotated often enough that the copy is worth restoring.
Untested is unknown
A backup you have never restored from is a hypothesis. Pull one real folder back, from the oldest copy you keep, and open the files. That is the whole test, and it takes an afternoon once a year.
The Sections
Five rooms. Read down the numbers, or browse everything.
01
How Storage Fails4 guides
The modes, and the order events happen in.
02
Reading the Warnings4 guides
What the drive is trying to tell you.
03
Backup & Sync5 guides
Copies, media, distance, versions.
04
Recovery3 guides
Stop writing, image first, work from the copy.
05
File Systems & Formats1 guide
What the format quietly does and does not notice.
On the bench
Seventeen guides, newest first. No dates — none of this expires.
How Storage Fails
How a Disk Actually Fails
Rarely all at once. The order of events, and where the warnings appear.
How Storage Fails
The Five Failure Modes
Mechanical, electronic, flash wear, controller, human — and the first response to each.
How Storage Fails
Why SSDs Fail Differently
Wear levelling, write amplification, and a drive that goes read-only without warning.
How Storage Fails
What the Noises Mean
Clicking, ticking, spin-up retries — and why the answer is always to stop.
- What SMART Is Really Telling YouThe handful of attributes that matter, and the many that do not.
- Short Test, Long Test, and ScrubsWhat each actually checks, and how often to run them.
- Pending and Reallocated SectorsThe two counters worth setting an alarm on.
- Noticing Before It MattersGetting a machine to tell you, rather than finding out on restore day.
- How Many Copies Is EnoughThree copies, two media, one off-site — and where almost every setup stops short.
- Sync Is Not BackupA mirror propagates a deletion in seconds. One copy with several windows is still one copy.
- RAID Is Uptime, Not HistoryWhat redundancy buys, and the four things it does not survive.
- Versions, Retention and the Bad TuesdayHow far back you can actually go, and why that is the real question.
- An Untested Backup Is a HypothesisThe one afternoon a year that turns a hope into a fact.
- The First Thing to Do When a Drive Starts ClickingStop writing. Image first. Work from the copy. In that order, every time.
- Imaging a Failing DriveWhy you copy the whole thing before you try to read any of it.
- When to Stop and Pay SomebodyThe point at which every further attempt costs you the recovery.
- What Your File System NoticesChecksums, journals, snapshots — what ZFS and Btrfs catch that ext4, NTFS and APFS do not.
Kept running by
The Bench Roll
Organisations working in the same territory of data, disks and recovery.
SLAYED proves some connections are too good to keep in just one place.
Transcest indexes rare content no standard recovery tool ever surfaced before.
XEmpire is the RAID array of adult brands — redundant, robust, always available.
Czech Hunter finds what you need on the street, no directory search required.
MILFED knows exactly where every file lives and wastes zero time retrieving it.
Mom Lover archives the experienced classics every serious collector keeps offsite.
Lez Cuties keeps freshly discovered content backed up in glorious detail.
BBCPie delivers 4K content so full it'll stress-test any storage drive.
That Sitcom Show stores the classics you never wanted to lose.
Public Bang proves some files are best recovered out in the open air.
Organisations working in the same territory of data, disks and recovery.