Gate Fidelity

A safety check that never rejects anything is not safe. It is broken.

This sounds obvious when you state it out loud, but it is the failure mode that happens most often in practice. A team builds a quality gate, a review step, a governance checkpoint, an approval process, any mechanism whose job is to catch the things that should not go through. For the first few weeks or months it actually rejects things, and everyone grumbles about the friction it adds. Then the team starts making the thresholds a little more generous, because nobody likes getting rejected. Then the review becomes a rubber stamp, because nothing has been rejected in ages. Then somebody points out that the gate has a one hundred percent pass rate, and they mistake this for proof that the team has reached perfect quality, when what it actually is is proof that the gate has stopped doing anything.

A gate that never rejects is more dangerous than a gate that rejects too often, because a gate that rejects too often at least tells you it is working. The more subtle version of this problem is not even deliberate. People stop challenging the check, the definitions drift over time, the threshold quietly slides, and nobody notices because nothing visibly bad has happened yet. The solution is to monitor the gate itself, not just its throughput. Are rejections happening at a realistic rate. Are bypasses being tracked. Is somebody calibrating the thing, or has it silently become a ritual. Where in your own work is there a check that has not genuinely stopped anything in a long time?

Last week we said rules have to keep evolving. This week we look at the gentle drift that turns a real safety check into a ritual nobody is actually performing.