A check that has never failed hasn't been tested

On watching your own safety nets catch something on purpose.

A note on one of my projects said a fix would make a check pass. The check did pass. It had also passed before the fix, for the same reason: it didn't test the thing the fix changed. Nothing in it looked for that field at all.

Nobody had done anything careless. The check existed, it ran, and it was green. Green is the colour that makes you stop looking.

A passing check tells you one of two things: the problem is gone, or the check can't see it. From the outside they look the same.

The only way to tell them apart is to break the thing on purpose. Cause the exact condition the check is meant to catch, watch it go red, then put everything back and confirm it goes green again. If you can't make it fail, you don't have a check yet. You have a line in a log that always says the same thing.

In electronics this is ordinary. You don't trust a fuse you've never seen blow, and you don't trust a meter you haven't put on a known source. Software lets you skip that step because nothing smokes when a test is wrong.

So I've made it a rule. Every new check gets tested red before it's trusted green, and the restore gets compared byte for byte, so the test doesn't leave a mark of its own. It costs a few minutes. The alternative is a safety net that only shows it's broken when something falls through it.

Back to the notebook