UTC(k), and who defines yesterday
UTC itself is virtual — computed after the fact, ticking nowhere, so it cannot be the thing you steer to. What you steer to is a realization: one institute’s live, physical best guess at UTC, running in hardware you could in principle go and look at.
These are written UTC(k), where k names the institute — UTC(NIST), UTC(USNO), UTC(PTB), UTC(NPL). Each is a real clock ensemble producing a real signal, right now.
The relationship, and how you learn it
Each institute steers its realization to track UTC as closely as it can, and each month Circular T publishes how well it did: the value of [UTC − UTC(k)] at five-day intervals.
So the chain is:
- Your clock follows some UTC(k), directly or through a distribution protocol.
- That UTC(k) follows UTC, imperfectly, in real time.
- A month later, you can look up exactly how imperfectly.
Every step of that is knowable. None of it is knowable at the time, which is the recurring shape of this whole topic.
Not every contributor is an NMI
A useful correction to an assumption people make: BIPM’s own word for the contributing institutions is “institutes”, not NMIs. Some are national metrology institutes, some are observatories, some are designated laboratories. UTC(USNO), for example, comes from the US Naval Observatory, which is not the United States’ NMI — that is NIST, which separately maintains UTC(NIST).
Who defines yesterday
Here is the part that surprises people. The clocks in the ensemble are not weighted equally. BIPM weights them by demonstrated stability, so a laboratory with better clocks pulls the average further toward its own reading.
Which produces a genuinely odd consequence:
The winners define exactly when yesterday was.
The institutes with the best clocks do not merely measure UTC more accurately than everyone else. They contribute more to what UTC turns out to have been. Their reading is more nearly correct partly because correctness is defined, in part, by their reading.
It sounds like a paradox and it is not. Weighting by stability is what makes the ensemble more reliable than any member — an unstable clock that got equal say would drag the average around. The consequence is simply that authority follows performance, continuously and automatically, with nobody voting on it.
What this means for you
Pick your UTC(k) deliberately. If you are steering to a national laboratory, its published [UTC − UTC(k)] history tells you what you are inheriting. They are not all equally close, and the record is free.
A GNSS constellation is a realization too, of a sort — one steered by its operator rather than by a metrology institute. That is the subject of the next page.
The offset you inherit is common-mode. Every clock in your estate that follows the same UTC(k) inherits the same error, which cancels in every measurement you make between them and survives in every measurement you make against anybody else.