The unit has named the six things it wants to watch in real time. It has also said, in public, that it cannot publish them yet

Monitoring something in real time is an engineering claim, not a figure of speech. It means data arriving fast enough to act on before the thing you are watching has finished happening. So it is worth walking the pipes that would carry it.

Britain has three main feeds on employment, and they run at different speeds.

The fastest is PAYE Real Time Information. Every time an employer runs payroll they file with HMRC, which means the state knows within days how many people were paid and roughly what they earned. That is genuinely close to real time and it is the best thing in the system. What it cannot tell you is why. A payroll file records that a person left. It does not record that a machine now does what they did.

The second is the Labour Force Survey, which asks households about work. It carries the detail payroll cannot: occupation, hours, whether someone is looking. It is also a survey with known response problems, currently published with a warning attached, and it reports on a rolling three month window. By the time a change is visible in it, the change is at least a quarter old.

The third is the Business Insights and Conditions Survey, which asks firms directly and every fortnight, and which since 2023 has asked about AI. It is the only feed that connects the technology to the work. It is also voluntary, excludes six sectors including finance, and asks businesses to describe their own behaviour.

Now the gap. To watch AI affecting jobs in real time you need something that links a specific technology to a specific change in specific work, fast. Payroll has speed and no cause. The household survey has cause and no speed. The business survey has both in weak form, twice a month, from firms rating themselves.

None of them was built for this question, because when they were designed the question did not exist.

What is unusual, and what this paper missed until last night, is that the unit has already written this down. The final section of its January assessment names six families of indicator it wants to track: how fast AI capability is advancing, how far the technology has spread through business, how jobs and workplaces are being reshaped, who is affected and where, the wider macroeconomic picture, and what new roles are appearing.

Read as engineering, that list is the specification for an instrument Britain does not currently have. Four of those six cannot be answered by any existing feed at any speed. Nothing in the national statistical system tracks the emergence of new occupations. Nothing links adoption at a firm to task change inside it.

And the assessment says so, in its own terms. The metrics are under development. The examples given are illustrative. Publishing updates on them is described as subject to agreement across government on the approach to monitoring and publication.

That last clause is the one worth sitting with. It is not a technical constraint. It says that whether Britain gets to see this, and when, is a decision still to be taken by departments rather than a capability still to be built.

So the honest reading of real time, on the unit's own published account, is an intention with a dependency attached and no date on it. That is a very ordinary position for a body eight months old, and it is a long way from what the phrase means in a chamber. Both of those things can be true, and this week I have been reminded that the second does not entitle anyone to assume the first.