I once deleted two hundred tests in an afternoon. Quality went up the same week.
I did not mark them for refactoring. I deleted them, with no ceremony. I had inherited a suite from a team that spent three years adding tests and eighteen months quietly muting the ones that failed for reasons nobody had time to chase. By the time I arrived, a fifth of the suite was switched off and still counted on the coverage badge.
Michael Bolton's distinction between testing and checking explains what had happened. Checking confirms what you already believe; testing investigates what you don't. Those 200 tests had stopped doing either. They were decoration wearing a green tick.
It is a smoke detector with a carrier bag taped over it because it kept going off at 3am. Still bolted to the ceiling, still technically installed, functionally just a bag.
Deleting them was the easy part. Admitting we had been lying to ourselves for eighteen months was harder.
What we did instead, and kept doing:
- Every skipped or muted test gets an expiry date, not an indefinite pass, and a name attached to it.
- If nobody can explain why a test is muted within five minutes, it's deleted, not investigated, not ticketed, deleted.
- The remaining suite gets read line by line once a quarter, not just run. Reading finds the theatre; running just confirms the theatre still executes.
- A shrinking suite is treated as evidence of trust, not evidence of neglect.
Within a week the build got faster, which was expected. Less expected, people started reading failures again, because a red build finally meant something.
Two hundred fewer tests did not make us less safe. It stopped 200 green ticks lying to us about it.
into your process?
A QA Review finds the gaps in 1–2 weeks. From £1,200.