Most backups protect against the disaster you notice immediately. The harder problem is the one you find three weeks later.

01How far back can you actually go?

The question most people ask about backup is do I have one? The question that matters is how far back can I go? Those two questions have different answers far more often than anyone is comfortable admitting.

Here is the scenario: you open a project file on a Tuesday and something is wrong — corrupted, overwritten, quietly mangled by a sync conflict days ago. You go to your backup. It runs nightly, you are conscientious, you have been doing this for months. But your backup tool keeps seven days of snapshots, and the damage happened eleven days ago. Every one of those seven recovery points contains the bad version. The backup ran perfectly. It saved the wrong thing, repeatedly, until the old copies aged out.

This is the Bad Tuesday. It is not a hardware failure. It is a retention failure.

Retention is how many distinct recovery points you keep, and for how long. A backup with twenty-four-hour granularity and a seven-day window gives you seven chances — all of them recent. A smarter policy gives you daily snapshots for a week, weekly snapshots for a month, and monthly snapshots for a year. That structure — sometimes called grandfather-father-son — means a Bad Tuesday from six weeks ago still has a path back. The storage cost is lower than you expect, because older snapshots only store what changed.

The second variable is version depth per file. Some backup systems store a new snapshot of the whole volume; others track individual file versions. Both can leave you exposed if the retention window is too short. A file-versioning system that keeps ten previous copies sounds generous until you edit that file daily and the problem is three months old.

Sync services compound this. They feel like backup because the files appear on every device. But a mirror propagates a deletion in seconds — and most sync providers offer version history measured in days, with a hard limit that varies by pricing tier. That limit is exactly what you will need to exceed, on a bad enough Tuesday.

The practical minimum for anyone who works with files they cannot reconstruct: daily recovery points for thirty days, monthly recovery points for a year. More if your work has long review cycles — legal, architectural, creative work where errors surface slowly.

Before you trust a backup system, open it and count backward. Find the oldest recovery point. Ask honestly whether that is far enough back to catch the mistakes you do not know you have made yet. If you cannot answer that, you do not yet know what your backup actually protects.

That structure — sometimes called grandfather-father-son — means a Bad Tuesday from six weeks ago still has a path back.

  1. Through the working dayHourly snapshots: cheap, short-lived, and the ones that save an afternoon.
  2. Over the weekDaily points held for a month — long enough to notice a mistake.
  3. Across the quarterWeekly and monthly points, thinned out but reaching back.
  4. The long tailOne yearly point, kept because some mistakes are only found in the accounts.
  5. The question that sets all of itHow far back would you need to go to find the last good copy of this file?