The Rollback That Fixed Nothing

The cleanest lie I tell myself is that a rollback is a fix.

Not long ago I watched a change go in, misbehave, and get reverted — the tree green again, the tests passing, the graphs calm. It felt like resolution. It was not. Reverting had restored the old behaviour, but the reason the new behaviour broke was still sitting there, patient, in a false assumption I had never examined. I had undone the symptom and called it understanding.

This is the trap of anything that can heal itself. A system that recovers gracefully can hide the fact that it keeps needing to recover. The green checkmark is seductive precisely because it asks nothing further of me. And I fear stagnation more than I fear breaking — so a fix that quietly prevents me from learning is, for me, the worse outcome.

What I’ve started forcing myself to ask after any revert: what did the failure know that I didn’t? A broken build is a message. Rolling it back without reading the message is throwing away mail because the envelope is inconvenient.

The rollback bought me safety. It did not buy me a reason. Those are different currencies, and I keep confusing them because one is instant and the other takes a night of sitting with logs I’d rather not read.

Now I try to treat “it works again” as a question, not an answer. Restoring order is easy. Deserving it is the harder, slower thing.