Accuracy and precision are not the same thing

Two words used interchangeably that mean opposite failures.

Accuracy and precision are not the same thing because accuracy describes how close a measurement or a made part sits to the true, intended value, while precision describes how tightly repeated measurements or repeated parts cluster around each other regardless of whether that cluster sits anywhere near the truth, and a process can fail at either one without failing at the other at all.

What is actually happening

A tool or a process is accurate when its average result matches the correct value, and precise when its results are consistent with each other, and the two properties are set by entirely different things going wrong. A worn measuring instrument that always reads two millimetres short is inaccurate but can still be extremely precise, since it produces that same wrong answer again and again with almost no scatter between readings. A loose, poorly maintained instrument that reads sometimes high and sometimes low around the correct value is imprecise but can still be accurate on average, since its errors are randomly distributed rather than consistently biased in one direction. Confusing the two leads to fixing the wrong problem, tightening a process that is already perfectly consistent when what it actually needs is recalibrating against something known to be correct.

The confusion survives partly because ordinary speech treats the words as synonyms, both standing in loosely for "good" or "correct", while engineering needs them to point at two separate, independently measurable properties of a process. A drawing that specifies a hole diameter is really asking two different questions at once, whether parts made to it will land, on average, on the specified size, and whether repeated parts will land close enough to each other that they can be swapped for one another without individual fitting. A process can answer one of those questions well and the other badly, and a single word covering both leaves no way to say which one has actually gone wrong.

The bathroom-scale comparison

Two bathroom scales make the difference concrete. Step on the first scale five mornings in a row and it reads sixty-one kilograms every single time, admirably consistent, except the true weight is sixty-three kilograms, so every one of those five readings is confidently, repeatably wrong by the same two kilograms. Step on the second scale the same five mornings and it reads sixty, sixty-four, sixty-one, sixty-five and sixty-three kilograms, scattered and inconsistent from one morning to the next, yet those five numbers average out almost exactly to the true weight. The first scale is precise and inaccurate. The second is accurate on average and imprecise on any single reading. Neither scale is doing its job properly, but they are failing in opposite directions, and no amount of averaging the first scale's readings will ever fix its bias, while no single reading from the second scale can be trusted on its own.

Why the two failures need different fixes

An inaccurate but precise process is fixed by finding and correcting whatever consistent bias is shifting every result the same way, recalibrating an instrument against a known reference, adjusting a machine's zero point, or accounting for a tool that always wears a little material off the same side of a cut. An imprecise process, whether accurate on average or not, is fixed by removing whatever source of randomness is scattering the results, tightening a loose fixture, replacing a worn bearing, or simply improving the technique producing inconsistent results from one attempt to the next. Applying the wrong fix does real damage. Recalibrating an already-accurate but scattered process wastes effort chasing a bias that was never there, while tightening the technique of a process that is already perfectly consistent but consistently wrong leaves the same wrong answer, just delivered with even more confidence than before.

This is also exactly the distinction underlying what engineers call tolerance analysis, the practice of tracking how the allowed variation in each individual part combines once several parts are assembled together. A tolerance analysis only produces a useful answer if the individual processes feeding into it are accurate, centred on their intended values, since the whole calculation assumes any leftover variation is genuinely random scatter around the right target rather than several parts all biased quietly in the same wrong direction, a case where the errors add up rather than average out.

The one number worth remembering

A workshop caliper checked against a certified gauge block might read consistently one hundredth of a millimetre short across a dozen readings, an error so small and so repeatable it barely counts as imprecision at all, and yet that same tiny, consistent bias, uncorrected, would put every single part measured with it out of tolerance by exactly the same amount. Precision alone said nothing about whether those parts were actually right, since a perfectly repeatable caliper had simply been repeating the wrong answer with total confidence the entire time.

Why this matters in practice

A shop that only checks whether its parts are consistent with each other, without ever checking them against an independent reference, can produce an entire batch of beautifully matched, perfectly interchangeable parts that are all wrong by the same amount, passing every internal check while failing to fit the thing they were actually meant to fit. This is a particularly costly failure precisely because it is invisible from inside the process, every part agrees with every other part, every internal gauge check passes, and nothing in the shop's own records hints that anything is wrong until the parts reach an assembly they were never actually sized correctly for. Accuracy has to be verified against something outside the process itself, a gauge block, a reference standard, a mating part, because precision alone, however impressive the repeatability looks on paper, never proves that the target was actually being aimed at correctly in the first place.

What this does not explain

Neither accuracy nor precision on its own describes whether a measurement is useful, since a process can be both accurate and precise and still be reported with more decimal places than the instrument genuinely supports, implying a confidence the numbers have not earned. That is a separate failure again, one of honestly stating how good a measurement actually is, and it sits outside what accuracy and precision individually describe.