A type B thermocouple barely cares what your cold junction is doing
Its curve doubles back near room temperature, so below 250 °C no inverse exists at all — and the same property makes a 40 °C swing in the cold junction move a 1018 °C reading by 0.05 °C.
·Source: NIST ITS-90 reference functions (SRD 60), cross-checked against Monograph 175
Put 0.15 mV from a type B thermocouple into a converter and a good one will refuse. Not warn — refuse. It is easy to read that as a gap in the tool, so it is worth saying plainly: below 250 °C there is no answer to give, and the reason turns out to be the same reason type B is used at all.
The curve doubles back
A thermocouple's emf is supposed to rise with temperature. Type B's does not, near room temperature. Evaluating the ITS-90 reference function for type B from 0 °C upward, in microvolts:
| T | emf |
|---|---|
| 0 °C | 0.00 µV |
| 10 °C | −1.88 µV |
| 21 °C | −2.58 µV |
| 30 °C | −2.12 µV |
| 40 °C | −0.49 µV |
| 42 °C | −0.03 µV |
| 50 °C | +2.28 µV |
It goes negative, bottoms out at −2.585 µV around 21.0 °C, and crosses back through zero near 42 °C. So −2 µV is produced at about 10.9 °C and at about 31.0 °C. One voltage, two temperatures.
That is not a measurement problem or a fit problem. An inverse function cannot exist where the forward function is not one-to-one. NIST publishes type B's inverse from 250 °C up, and a converter that answered below that would be inventing a branch.
The other seven letter types rise monotonically across their whole span, which is why only type B behaves this way.
The same shape is why type B exists
A thermocouple measures the difference between its two junctions, so every reading needs the cold junction's own emf added back before it can be inverted. Getting that wrong, or letting the terminal block drift, is the classic source of error.
Now watch what a cold junction does to type B. The same measured 5.000 mV, with the cold junction sitting anywhere from freezing to a hot cabinet:
| Cold junction | Its contribution | Hot junction reads |
|---|---|---|
| 0 °C | 0.00 µV | 1018.034 °C |
| 20 °C | −2.58 µV | 1017.755 °C |
| 25 °C | −2.49 µV | 1017.764 °C |
| 30 °C | −2.12 µV | 1017.805 °C |
| 40 °C | −0.49 µV | 1017.981 °C |
A 40 °C swing in the cold junction moves the reading by 0.05 °C. The contribution never exceeds three microvolts because that is the entire height of the dip — the curve simply has nothing to contribute down there.
Type K, for contrast, on the same 5.000 mV:
cold junction 25 °C → 146.60 °C
Twenty-five degrees of cold junction moves a type K reading by 24.65 °C — very nearly one for one, which is what a well-behaved thermocouple does. Type B's flat spot near room temperature is a liability below 250 °C and an asset above it.
What this means in practice
Cold-junction compensation still belongs in your code. The error is small, not zero, and at 0.05 °C per 40 °C it is free to get right. What you can stop worrying about is cold-junction sensor accuracy and terminal-block gradients — the things that dominate a type K budget barely register here.
Do not use type B below 250 °C. Not because the tooling refuses, but because the sensor produces microvolts there. At 100 °C type B makes about 33 µV, where type K makes 4.096 mV — two orders of magnitude more signal. Any noise on the leads swamps it.
A reading near zero is ambiguous, not wrong. If a type B loop reads a couple of microvolts, that is consistent with a cold process, a disconnected sensor, or a junction at 31 °C. You cannot tell from the voltage. Type B is a high-temperature instrument and its behaviour at the bottom carries no information.
Why converters should refuse rather than extrapolate
It would be easy to return the lower branch, or the upper one, or to extend the 250 °C polynomial downward and let the reader decide. All three produce a number that looks like a measurement.
The honest output is the range the inverse is actually published for, and the reason. A refusal costs somebody thirty seconds of confusion. A plausible number from a region where the physics does not support one costs whatever it was controlling.
The same principle covers the wire colours, which this calculator also does not show: ANSI/ASTM E230 and IEC 60584-3 assign different colours to the same type, so publishing one set would mislead whoever is holding the other cable.