When I walk a client through incident response, I usually pick ransomware, because it forces the whole chain. Pull the affected machine off the network and leave the power on, since powering down loses the memory-resident evidence and a reboot can trigger the second stage of some ransomware. Check the logs to see what left the network. Rebuild the machine. Then restore from backup.
That last step is where I say something plainly. If you can show me 6 months of consistent restore testing under your own backup policy, followed to the letter, I'll buy you a coffee. I'm that confident most places don't have it.
Backups are a belief until you restore one
Most shops have a backup job. It runs, it turns green, and somebody gets an email. That tells you the job ran. It doesn't tell you the data comes back, comes back complete, or comes back in a usable time.
The only evidence that a backup works is a restore. The only evidence of a restore, to an assessor or to yourself in 18 months, is a record that it happened.
The log most shops can't produce
The restore test log in my templates has 7 columns: date, system, restore point, method, result, time to restore, and who tested. The sample rows are a full VM restore of a file server and a file-level restore from a project share, with times next to each. Under the table goes the date of the next test and the backup policy the tests are measured against.
No tooling. A spreadsheet in the evidence folder is enough.
What makes it evidence is repetition and variety. 1 test proves it worked once. A quarterly cadence across different systems and restore methods proves the process runs, and it shows you drift. If the file server restores in 42 minutes in September and takes 4 hours in December, you learn that before you need it.
This is the same test that applies across a whole compliance program: the policy says it happens, the log shows it did, and the gap between the two is where assessments fail.
Write a policy you will follow
The coffee line has a trap built in: "under your own backup policy, followed to the letter." Most backup policies are aspirational. Daily offsite copies, quarterly restore tests, monthly verification. Then reality arrives, and the policy describes a company that doesn't exist.
My template policy is 3 copies, on 2 media types, with 1 offsite. That's 3-2-1. Write down which systems it covers, how often each copy is made, and how often you test, then pick a frequency you can sustain. A quarterly test that happens beats a monthly test that was done twice and abandoned.
For a CPCSC shop, 3 specifics matter.
- Backups are in scope. A backup of the drawings is a copy of Specified Information. Give the enclave its own backup job, and treat the backup server as inside the boundary.
- Backups are encrypted. 03.13.08 requires encryption of Specified Information on servers, laptops, backups, and portable media alike.
- Someone owns it. If the MSP runs the backups, the responsibility matrix should list the restore test log as an MSP deliverable, quarterly, with the client keeping a copy. If it doesn't arrive, that is a finding about the MSP, and you want to see it first.
Restore before you reconnect
The recovery step in my incident response skeleton reads "restore from tested backups; verify before reconnecting." Tested is the operative word. A real incident is the worst time to learn how long a restore takes.
Run the first test this month. Pick the system you would miss most, restore it to a spare location, time it, and write the row. If it fails, you found the problem on an ordinary day, which is the cheapest day to find it.

