Britain's main measure of its own workforce comes with a warning label from the people who write it. The part it sees worst is the part changing fastest

Every argument we have about AI and jobs assumes someone is keeping score. That when the machines start thinning the work, we will see it in the official figures: the unemployment rate ticking up, vacancies falling, a line on a chart a minister has to explain. The assumption is that there is a referee, and the referee can see the pitch.

Read the fine print on Britain's labour figures and that assumption does not survive.

The unemployment rate, the employment rate, the number of people in work that leads the news: nearly all of it comes from one source, the Labour Force Survey, run by the Office for National Statistics. It has been the country's main read on its own workforce for decades. And for the past two years it has carried a label, in ONS's own words: not "official statistics," but "official statistics in development." The most recent bulletins say it plainly. They advise caution when using the data, and say the figures should be read alongside other sources, because of ongoing problems with how many people respond.

Here is what happened, without the euphemism. In the middle of 2023 the share of households answering the survey fell to a low, under thirteen in every hundred approached. For a stretch that winter, ONS stopped publishing the detailed figures altogether, judging them not reliable enough. It started again in early 2024, but rebadged them as "in development," a health warning it has not yet lifted.

The honest part, which the reassurance merchants tend to skip: response has recovered a lot since. More interviewers, a bigger sample. That is real and it matters. But a recovered top line is not the same as a clear picture underneath. The regulator that oversees official statistics noted this spring that the sample still does not stretch to detailed regional or devolved analysis. The survey can tell you roughly how the whole country is doing. It struggles to show you the corners.

And the corners are the story. This morning this paper reported the ONS firm-size figures: vacancies down around 15% in the smallest firms, and holding, even edging up, in the giants. The loss is concentrated at the bottom, in the small employers and the first jobs, the places the national instrument sees least well.

There was meant to be a fix. A new, online-first survey, the Transformed Labour Force Survey, with a far larger sample, built precisely to see more clearly. It has slipped, and slipped again. ONS once hoped to move the headline figures across to it late this year, after a readiness check this month. In April it admitted it would not have the data it needs for that check. The earliest the switch now happens, the regulator says, is 2027. Running two surveys at once in the meantime is straining the resources of the body doing it.

So line it up. The tool is labelled handle with care. The part of the labour market moving fastest is the part it sees worst. And the better tool is years away. None of this is hidden. It is written down, by the statisticians themselves, for anyone who reads to the end of the methodology note. Which is the quiet scandal of it: we are holding a national argument about a thing we have, on the record, half-decided not to be able to see.