Explainable by default

Most maintenance tools work like a magic trick. You press a button, a progress bar fills, and some number of gigabytes disappears. It feels good for a second, and then you wonder what exactly just left your disk.

Atlas is built to remove that second feeling. It runs under a single constraint: every action is explainable before it runs, and recoverable after it does.

Explainable before it runs

Before Atlas changes anything, it shows you the change. Not a category — the actual files, caches, and logs it proposes to touch, with the reason each one is on the list. You read the plan, then you decide. A tool that cannot say what it is about to do has not earned the right to do it.

This rules out a whole class of convenient behavior. Atlas will not "optimize" in the background, and it will not bundle a dozen unrelated cleanups behind one reassuring button. Each step stands on its own and states its case.

Recoverable after it does

The second half matters as much as the first. Cleaning up should not be a one-way door. When Atlas removes something, it does so in a way you can roll back, so a cleanup that looked right on Tuesday and wrong on Wednesday is a reversal, not a restore-from-backup afternoon.

Recoverability changes how the tool feels to use. You stop bracing before you run it. The stakes of any single action drop to near zero, because any single action can be taken back.

Why open source follows from this

A tool that asks you to trust its explanations should let you check them. Atlas is open source for that reason, not as a marketing note. If we claim an action is safe and reversible, the code that makes that claim is there to read. The promise and the proof live in the same place.

That is the entire design in one line: see it before it happens, undo it after. Everything else in Atlas is downstream of holding that line.