Monitoring that publishes only when everyone agrees is not monitoring. It is a negotiation.

There is one clause in the January assessment I cannot stop looking at, and it is not a finding. It is a condition.

Subject to agreement across government on the approach to monitoring and publication, the Unit intends to publish updates on these indicators.

Read it as an ordinary sentence and it is unremarkable. Cross-government work needs cross-government agreement on method. Statisticians are careful people and they ought to be.

Read it as a description of an incentive and something else appears.

The six indicator areas that clause governs include distributional impacts, outcomes for early-career workers against experienced ones, and the macroeconomic picture including the tax base. Those are the measurements most likely, on some futures, to produce a number nobody in office wants published. A conditional publishing arrangement is fine while the readings are neutral. Its properties only become visible when a reading is bad.

That is the behavioural point, and it is not a claim about anybody's motives. Nobody has to intend suppression for this to bite. Agreement across government is simply harder to reach about an uncomfortable number than a comfortable one, and the delay that produces is indistinguishable from ordinary care. A body publishing on its own authority ships when the data is ready. A body publishing by consensus ships when the last department is content. Those two arrangements produce identical outputs for years and then diverge at exactly the moment the output matters.

There is a second thing worth noticing about the same document, and it runs the other way, so I will give it its due.

The assessment does not flatter its own case. It sets out the evidence that hiring has fallen faster in AI-exposed occupations, then dismantles the causal reading itself, offers interest rate tightening as a rival explanation, and imports two studies that find no effect at all. It says out loud that the available evidence does not answer the questions that matter most for policy. That is not the behaviour of a body building a case. It is the behaviour of one trying to see straight.

Which is what makes the placing question sting. The instrument looks like a good one. Careful, self-critical, willing to publish its own uncertainty.

A good instrument you cannot locate is not much use.

I want to be careful about one thing before I finish. Somebody reading this could conclude that the Unit is being suppressed. Nothing here supports that, and I do not believe it. The clause describes a structure, not a plot, and structures can be fixed with a sentence. If the government wanted to close this question tomorrow it could name the Unit's new home and say that its indicator publications will go out on its own timetable. Both of those are free.

What we have instead is a body with a conditional licence to publish and, as far as the paperwork goes, no address.