Mother's Day: "did you check it twice?"

Happy Mother's Day to everyone celebrating today, and to everyone missing someone today too.

One of us was on the phone with their mom this morning. Somewhere between the flowers arriving and a long story about the neighbor's new dog, she asked the question she has asked for as long as anyone can remember:

"Did you check the stove before you left?"

Not "did you turn it off." Did you check. And then, before the answer was even out: "Did you check it twice?"

It's a family joke now. Growing up, leaving the house meant a little ritual. Stove knobs, all pointing the same way. Iron unplugged. Front door locked, then the handle tried again, just to be sure. Once you'd done it, you'd done it. But you'd do it once more anyway, because the first time you were thinking about your keys.

We laughed about it on the call. She pointed out, fairly, that in thirty-some years the house has never burned down. Hard to argue with a perfect record. Then, as these things go, we ended up talking about it at work.

Because that's more or less what we're designing our referee to do. When an explorer agent thinks it found a bug, we don't take its word for it. A separate process replays the exact steps in a clean browser session, with no model involved, and we ask it to see the problem at least twice out of three tries. For findings that only an agent noticed, we ask for all three. We wrote about the math behind that in Pi Day math: why we replay three times and accept two.

The reason is the same as the stove. The first look is unreliable, not because anyone is careless, but because attention is split. The agent was thinking about its charter, the page was still loading, a banner was halfway through animating. Looking again, calmly, on purpose, separates "I think so" from "yes."

Moms figured that out a long time before anyone wrote a paper on flaky tests.

So if you're calling home today, maybe tell her she was right about the stove. And then check it twice.