Keep calibration evidence in a fair comparison

Treat readings as evidence, not a verdict

Suppose two prototypes differ by a few millimetres and must fit the same enclosure. More displayed digits might help distinguish them, but those digits cannot by themselves establish accuracy. Before using a measurement to pass a design, investigate what the instrument and procedure can support.

This lab uses an exact 10 cm reference. Its offset control supplies a known additive error between −2 and +2 cm, in 0.1 cm steps. Its resolution control selects a displayed increment from 0.1 to 1 cm. The three deviations, −0.2, 0 and +0.2 cm, are programmed teaching inputs. They are not random samples or measurements you made with a physical instrument.

Predict the raw readings first

Set offset to +1 cm and resolution to 0.5 cm. Before rounding, the three values are 10.8, 11.0 and 11.2 cm. Rounding each to the nearest 0.5 cm produces 11.0, 11.0 and 11.0 cm. Their mean is 11.0 cm. Identical displayed readings conceal the small input differences; they do not show that the reference became 11 cm.

The model supplies the offset, so subtract it: 11.0 − 1.0 = 10.0 cm. This corrected mean agrees with the modeled reference. We have not independently estimated an unknown instrument error or established a calibration chain. Keep the raw readings, the supplied correction and the corrected result in separate columns.

Now hold offset at +1 cm and change only resolution to 0.1 cm. The displayed values become 10.8, 11.0 and 11.2 cm; the mean remains 11.0 cm and the corrected mean remains 10.0 cm. Finer resolution reveals more variation. It does not remove the additive error. Capture both settings and label exactly which control changed.

Look for what remains after correction

Hold resolution at 0.5 cm and set the supplied offset to +0.3 cm. The values before rounding are 10.1, 10.3 and 10.5 cm. Displayed readings are 10.0, 10.5 and 10.5 cm, with mean 10⅓ cm. Subtracting 0.3 cm leaves approximately 10.033 cm. A small residual remains because rounding occurred before the correction.

Do not report the corrected mean’s extra decimal places as an accuracy certificate. This deterministic example explains an operation order. It does not provide an uncertainty distribution, confidence interval or physical instrument specification. A real reference also has uncertainty, whereas this teaching reference is exact.

NIST’s traceability policy emphasizes that traceability alone does not guarantee fitness for a particular measurement purpose. Here the narrower lesson is practical: state the question, correction and limits before deciding whether a reading is adequate for a design tolerance. The examples and calculations on this page are original hypothetical activities, not NIST calibration results.

Plan a fair prototype comparison

On paper, propose measuring both prototype thicknesses with the same reference check, instrument settings, contact method and stated conditions. Record the order, all readings and any reason for excluding a value. Repeatedly reading one unchanged part explores a different question from measuring independently manufactured parts. Neither automatically diagnoses a shared offset.

For thermal design comparisons, measuring thickness carefully is only one part of the method. Preserve area and boundary temperatures, identify the conductivity evidence and distinguish modeled from measured heat transfer. A calibrated ruler cannot repair unequal thermal conditions. Repeating a confounded comparison makes more records without necessarily removing the confounding.

Before using a correction over a wider range, check another known reference. One point can support a local additive correction while leaving a scale error undetected. The widget provides only the 10 cm reference; additional lengths in the following activity are paper evidence, not another implemented control.

Transfer: when one correction does not transfer

A hypothetical device reads 10.4 cm at a 10.0 cm reference and 20.8 cm at a 20.0 cm reference. A colleague proposes subtracting 0.4 cm from every reading. Predict the corrected second reference and identify a plausible alternative error model.

Check the two-reference transfer

Subtracting 0.4 from 20.8 gives 20.4 cm, still 0.4 cm too high. Both supplied values are consistent with a multiplicative factor of 1.04, so dividing by 1.04 fits these two exact paper points. They do not prove that this relation holds throughout the instrument's range. Additional references, uncertainty and stability checks would be needed. This public worked task is not independent assessment evidence.

Retain your matched-comparison protocol alongside the captures. Return to controls and replication or calibration and bias as needed. The useful result is a defensible measurement procedure and a bounded conclusion, not simply a corrected number.