The instrument that drifted

A school's weather station reported that the playground had warmed by 1.4 degrees in six years. The playground had not. The sensor had.

Drift is the slow, unintended change in what an instrument reports when nothing about the world has changed. Electronics age, surfaces discolour, mountings loosen, and the effect is almost always in one direction, which is what makes it dangerous. Random scatter announces itself: the readings jump around and anyone can see that a single measurement means little. Drift is quiet, consistent and looks exactly like a trend.

The defence is calibration against something that does not drift, or at least does not drift the same way. The school's eventual fix cost nothing: once a month, a student took three readings with a cheap alcohol thermometer in the same spot, and the difference between the two instruments was recorded. For the first two years the difference was within half a degree. After that it grew steadily, and the growth, not the temperature, was the finding.

Two further details matter. The comparison thermometer was stored indoors and taken out only for the reading, so its own ageing was slower. And the log recorded the difference rather than a corrected temperature, which meant that later students could see the evidence rather than a conclusion.

The habit generalises to anything measured repeatedly over years: a bathroom scale, a fitness tracker, a marking scheme. The question is not whether the instrument is accurate, which is often unknowable, but whether it is accurate in the same way it was last year. Only a second, independent instrument can answer that, and it does not have to be a good one.