The Word ID Got Typed Into the ID

Every lab note we publish carries a reference number, and the citation block at the foot of the page hands that number to anyone who wants to quote us. It is the one string on the page built to be copied somewhere else.

On 74 of those notes, the reference number had a form label typed into it. Where the field should have held the number alone, it held the letters ID, or a stray D, sitting in front of it. The citation block does not clean what it is given, so it published the whole thing, label and all.

Anyone who copied one of those citations copied a reference that resolves to nothing. The BibTeX version was worse, because its key is built from the same field with the spaces stripped out, so the label fused to the number and became part of the identifier.

Nothing about any product was wrong. No score moved, no price changed, and no claim about anything we tested is affected. What was wrong was our own name for our own record, which is a smaller thing to get wrong and a stranger one.

Why it lasted

The value was correct in every way our checks are able to ask about. The field existed, it sat inside its length limit, it was a string where a string belongs, and every rebuild of those pages copied it forward exactly as stored, which is what a faithful copy is supposed to do.

None of our checks read what is inside a field against the shape that field is meant to have. They check that fields are present, that they fit, that the right code reads them, and that the documents we publish agree with the code. A field holding the wrong contents passes all four.

What found it was not a check at all. We moved the reference number from a small row at the foot of the page to a stamp above the headline, and the moment it sat somewhere a person would actually look, it was obvious. The detector was a layout decision.

What changed

The label is gone from the field on all 74 notes. The number we kept is the number that was already sitting inside the field, so nothing was allocated, invented or renumbered, and every citation now resolves to the record it names.

We also found the same fault one level quieter. Another 282 notes carry a stray space around an otherwise correct number, which is invisible on the page and carried intact into the machine formats. That one is queued as its own fix rather than folded into this one.

On 20 August 2026 we found the same box publishing a number no field held, and on one note it had offered a confidence reading as our score.