Choosing calibration points

• Cover the range of use, including both ends; do not extrapolate beyond the calibrated points.
• Include the specific temperatures at which the instrument makes decisions — alarm limits, storage set points.
• Use enough points to reveal non-linearity; three is a minimum for a simple sensor, more for wide ranges.
• For PRTs, include or approach 0 °C where a known reference point exists.
• Match the medium and immersion to service conditions where practical.

A logger used only in a 2–8 °C refrigerator does not benefit from a certificate at 25 °C and 50 °C; it needs points at and around its limits.

Three questions that choose the points for you

Points are chosen from use, and the choice usually falls out of three questions asked in order.

Why you cannot extrapolate

A certificate establishes the relationship at the points measured and, by interpolation, between them. Outside them it establishes nothing, and this is not a technicality: a Pt100 with a well-behaved correction of +0.03 °C across 0 to 100 °C can be several tenths out at 250 °C for reasons — strain, encapsulation, lead effects — that are invisible from the calibrated range. A logger certified from 0 to 10 °C and used in a −20 °C freezer has an uncertainty nobody has evaluated.

The practical corollary: calibrate to the edges of use, not to the middle of it. A certificate at 2, 5 and 8 °C does not cover an 8.4 °C excursion, which is exactly the reading a decision will turn on.

How many points

• One point gives an offset and says nothing about slope. Adequate only where the instrument is used at essentially one temperature and its behaviour elsewhere does not matter.
• Two points give a slope but cannot reveal curvature, and every real sensor has some.
• Three to five is the usual industrial answer: both ends, the middle, and any decision temperature.
• More are needed where coefficients are being fitted rather than corrections tabulated. Fitting ITS-90 or Callendar–Van Dusen coefficients is a different and larger requirement than recording a few correction values.
• Match the points to the interpolation you intend. If the readout will apply a fitted curve, the points have to support the fit; if you will interpolate linearly between corrections, the points have to be close enough for that to be true.

A logger used only in a 2–8 °C refrigerator gains nothing from a certificate at 25 °C and 50 °C, and laboratories offer those points because they are convenient to generate rather than because you need them. Specify your points; do not accept the standard set.