[asveikau]

I think most of us have probably deployed fixes without knowing 100% whether it helps, but based on a good hunch from looking at data and code. The reason I keep doing so is because so long as the hunch is good and based on facts, it keeps working.

Tests are good, but there are always those issues those 0.001% issues that will never occur when you are trying to make them happen, but will happen to somebody once code is deployed. I am reminded of an incident a few years back where I could reproduce it with tests occasionally, but it would literally take me weeks of re-running the test 24/7 before I would hit it again. As you can imagine, I applied some "blind faith" (as in article) fixes while waiting for that stuff to run.

PS: By the rules of wikipedia, this article probably shouldn't exist. Nonstandard terms, not notable, original research, blah blah blah.