Measuring the same part twice gives two answers

Measurement uncertainty, and why repeat readings disagree.

Measuring the same part twice gives two slightly different answers because every real measurement carries some unavoidable variation, from finger pressure, from the angle the instrument sits at, from tiny changes in temperature between one reading and the next, and no instrument or operator repeats a measurement with perfect mechanical identity however carefully the second attempt copies the first.

Two pencil marks on the kitchen wall

Standing against a wall to have a height marked in pencil, then stepping away and standing there again a few minutes later, rarely produces two lines at the same height. Posture shifts, hair compresses differently under the book laid flat on top, and the person marking never holds the pencil at quite the same angle twice. Nobody concludes they grew or shrank in those minutes. Everyone understands that the true height stayed put and the measurement of it wandered a little.

A precision instrument on a manufactured part obeys the same principle at a much smaller scale. A measurement is the outcome of a physical interaction between the part, the instrument, the person or machine operating it, and the surrounding conditions, and each of those varies slightly from one attempt to the next. Caliper jaws never close with identical force twice, a part's temperature drifts by a fraction of a degree between readings minutes apart, and the angle of contact shifts by amounts too small to see. None of this means anyone did anything wrong.

Narrowing the scatter

Better technique, controlled temperature, training for consistent pressure and a more precise instrument all shrink the spread between repeated readings. None of them takes it to zero, because an instrument and a part meet under real force and both surfaces deform microscopically wherever they touch, an effect that belongs to the materials themselves. A measurement process is judged by whether it keeps the scatter comfortably small relative to the tolerance the part needs to meet.

This scatter is random, landing above and below the true value in roughly equal measure, so averaging helps. The random ups and downs partly cancel, and the spread of an average shrinks with the square root of the number of readings, so averaging four readings halves the scatter and averaging a hundred cuts it to a tenth.

Scatter and bias are different faults

An instrument or technique that reads consistently too high or too low behaves quite differently. Averaging its readings only produces a more confidently wrong number. Telling the two apart requires checking the process against a known, independent reference standard, since agreement between repeats says nothing about agreement with the truth. That standard is itself only as good as the chain of earlier calibrations behind it, which is why a credible process records when and against what its instruments were last checked.

Reporting a single bare number overstates how much it can be trusted, while reporting the spread of several readings alongside the average gives anyone relying on the result an honest sense of its confidence. The rest of this set follows how this scatter gets misread, mistaken for precision when an instrument shows more decimal places, or ignored when two parties bring their own honestly different readings of the same part to the same argument.