Visualizing precision, trueness, and accuracy

Everyone has seen the archery target. Almost everyone has seen it labeled wrongly, and the wrong label is not a pedantic quibble — it hides the single most useful fact about averaging.

The axes are trueness and precision. Accuracy requires both.

The usual version labels the rows “accurate”. They are not — they are true. Accuracy is the boxed corner where both are good, and calling one axis by the name of the whole thing is what makes “accurate to 15 ns” ambiguous.

Two things vary here and only two: how tight the group is (precision) and where its center sits (trueness). Accuracy is not a third dimension. It is the verdict when both are good.

Once the axes are named correctly, the thing everybody wants to know falls out of the picture: averaging moves you up the precision axis and never sideways along trueness.

This is the claim the four static quadrants cannot make, and it has a page of its own: the limits of averaging, where the same shape turns up in position, in time, and in frequency.

The panel that makes it honest

The shots are identical in both panels. Only the frame differs — and the frame is the thing you do not have.

Here is where the target quietly lies. On a real one you can see the bullseye. In measurement you cannot: you see your own results and nothing else, and from inside, a tight group offset by a meter and a tight group dead center are indistinguishable.

Worse in this field than in most, because UTC is a deferred-time scale — the center is not merely unobservable, it does not exist yet when you fire. It is settled afterwards and published in arrears.

The points are identical in both panels. Nothing about the clock changed — the target got painted. At the moment of measurement only the offset from UTC(k) is knowable; the red arrow, which is the trueness you actually care about, can only be drawn in arrears.

Which gives traceability its job. An unbroken, documented chain to a reference is the only thing that ever tells you where the center was.

See also: relating accuracy to resolution, and what happens when you would rather your clocks agreed than were right.