Read the label, not your memory

We wrote a number down wrong this week. Not a dangerous one: a count of our own automated tests, quoted in a release note as seven thousand three hundred and eighty when the real figure was seven thousand three hundred and seventy-four.

Nothing came of it and we corrected it the same morning. What makes it worth writing about is how it happened, because the mechanism is exactly the one that produces allergen incidents.

The true number had been on screen minutes earlier. It was read, it was understood, and then it was typed from memory instead of being read again. The remembered version was close enough to look right, which is precisely what made it dangerous: a wildly wrong figure gets caught, a plausible one gets published.

Close enough looks the same as right

A kitchen version. Somebody asks whether the burger bun contains sesame. The chef knows this bun. They have handled it a thousand times. They say no, and they are remembering last year's bun, because the supplier changed the spec in the spring and the box looks identical.

Nobody was careless. The chef was confident for good reasons, built up over a year of being correct. Confidence is what memory feels like from the inside, and it feels exactly the same whether the memory is current or stale.

Which is why the rule cannot be try harder to remember. It has to be: the person who answers the question reads the label.

Three habits that hold

The answer comes from the artefact, in the moment. Not from the person who knows, not from the folder as it stood at the last review. If somebody asks, the packet gets picked up, or the written record gets opened, and the answer is read off it while the guest waits. Fifteen seconds of waiting is not a service failure. A wrong answer delivered instantly is.

Write the claim next to what you read. When you record that a dish is free from something, record where that came from and when. A line that says checked against the box on Tuesday can be trusted on Wednesday and questioned in March. A line that just says no cannot be questioned at all, which sounds like strength and is actually the opposite.

A substitution is a new question, not a variation on an old answer. The wholesaler is out of the usual mayonnaise and sends something else. The dish is the same dish, the recipe card is the same card, and every allergen statement attached to it now refers to a product that is not in the building. Substitutions are where memory does the most damage, because everything visible is unchanged.

When two records disagree

The last one is about what to do when you catch it.

If the folder says one thing and the box says another, the box wins, but that is the easy half. The important half is that you now know one of your records drifts, and the fix is not to correct the single line. It is to ask how the folder got a year out of date without anybody noticing, and what else in it came from the same source.

We did that with our own slip. The wrong number went into a written record that exists specifically so nobody has to rely on memory, which is a small irony and a useful one: the record was there, it was open, and the shortcut still happened. So the correction names the mistake in the record itself rather than quietly replacing the figure. Anybody reading it later sees both what was published and what was true, and does not have to wonder which of the two they are looking at.

A written record that has been corrected in the open is more trustworthy than one that has never visibly been wrong. The second kind is not more accurate. It is just less honest about being a record kept by people.