Two thermometers, two answers: where the real temperature error is hiding

Infrared guns read a surface and probes read the bulk, and both drift. Ice point checks, emissivity error and immersion depth, with the tolerances a pour needs.

Two instruments in the same pot give two numbers, and the instinct is to assume one has drifted. Drift is rarely the culprit. In a small workshop the dominant errors are geometric: what the instrument is pointed at, how deep it is sitting, and how long it has been there. This page separates the errors you can check in five minutes from the ones you have to design out of your method.

• Working tolerance: plus or minus 1 C
• Ice point: 0 C within 0.3 C
• IR guns assume emissivity 0.95
• Immerse 10 to 15 probe diameters

They are not measuring the same quantity

An immersed probe measures the temperature of the material touching its sensor, by conduction, once it has equilibrated with it. An infrared thermometer measures the flux of long wave radiation arriving at its detector from everything in its field of view, and converts that flux into a temperature using two assumptions: an emissivity, usually fixed at 0.95 on cheap units, and a reflected ambient temperature.

Those are different quantities, and on a balm melt they have different answers even when both instruments are perfect. The surface of an open stirred melt is its coldest part, radiating to a 20 C room and losing heat to convection while the stir keeps renewing it, and the gap to the bulk commonly runs to several degrees. Much of what looks like instrument disagreement is two correct answers to two different questions.

That settles which one governs. Every decision containing the word "pour" is a decision about the bulk, because the bulk is what goes into the container, so the probe wins by definition rather than by being better made. The gun keeps its own job: whether a tin has set, whether a mould is still warm, whether the worktop is cold enough to shock the balm. That division of labour is the one set out in balm making equipment.

Start with the ice point, because it is free and it is definitive

An equilibrium mixture of ice and liquid water at atmospheric pressure sits at 0.00 C. That is a property of the substance, not an approximation, and done properly it reproduces to within a few thousandths of a degree, far better than any instrument you own can resolve.

1. Crush or shave the ice. Cubes are the commonest cause of a bad ice bath. You want particles, so the slurry has no pockets of water sitting away from an ice surface.
2. Fill an insulated vessel with ice, then add water. Only until it fills the gaps. The slurry stays mostly ice, and a spoon pushed in should meet resistance.
3. Stir, and keep stirring. Push the sensor at least 50 mm in, well away from the sides and the bottom.
4. Wait 30 seconds and read. The settled value, not the number it passes through on the way.

The mistake is a few cubes floating in a glass of water. Water is densest at about 4 C, so the warmest water in that glass sinks to the bottom, which is exactly where most people put the probe. That error alone reads several tenths high, occasionally more than a degree, and it has condemned a lot of perfectly good thermometers.

The boiling check, and why altitude is only half the correction

Water boils at 100 C at 101.325 kPa and nowhere else. Two things move it, but craft guidance mentions only one.

The working approximation is about 3.3 C per 1000 m of elevation, or 1 C per 300 m. The second correction is weather: near 100 C the boiling point shifts roughly 0.28 C per kilopascal, and station pressure at one place can swing four kilopascals between a deep low and a strong high, which is more than a degree of movement without going anywhere. Note the trap in looking that up, too. The pressure a weather app shows is corrected to sea level, so it is the wrong number unless you live there. You want uncorrected station pressure.

Treat the boiling check as worth about plus or minus 1 C. It is still worth doing, since it is the only high temperature reference most people can generate and span error only shows up away from zero. But there is a better one for this craft: stand a jug of oil in a water bath at 70 C, stir it, and compare every instrument against your reference in the range you actually work in. That check is described under weighing and calibration, and it interrogates the part of the scale where pour temperatures live.

Emissivity: why a gun aimed at a steel pan reads cold

Every surface radiates less than a perfect black body, and emissivity is the fraction it manages. Organic liquids including molten balm sit around 0.94 to 0.97, which is why fixed guns are set at 0.95. Bare metal is nothing like that: polished stainless around 0.1, brushed stainless 0.3 to 0.4, bright aluminium as low as 0.05. Aim a 0.95 instrument at a shiny pan and it sees a tenth of the radiation it expects, plus a reflection of the cool room.

Read the brushed stainless row and the argument is over. A jug genuinely at 70 C reads about 41 C, and that number looks like a melt needing more heat, which is how people end up holding fat above the 85 C ceiling discussed in melting methods while the gun reassures them.

Two remedies. Aim only at the liquid surface, never at the vessel and never where the spot can catch the rim. Or, for a repeatable reading off metal, stick a square of matt black electrical tape on the outside, let it come to temperature and aim at that: its emissivity is close to the assumed 0.95. Neither turns a surface reading into a bulk reading, so neither licenses a pour decision.

Distance to spot: you are averaging a circle, not reading a point

The laser dot is an aiming aid, not the measurement. The instrument integrates a circle whose diameter grows with distance, and the distance-to-spot ratio on the label says how fast. A 12:1 instrument at 300 mm averages a circle about 25 mm across, and 50 mm at 600 mm. An 8:1 budget unit at 300 mm averages 37 mm, wider than many pouring jugs.

Two consequences. The target must be comfortably bigger than the spot, and the working rule is at least twice its diameter, because the nominal figure is usually the circle containing about 90% of the energy and the rest arrives from outside it. And on a small jug the spot inevitably includes some rim, which is bare metal, so the low emissivity above drags the whole average down. A reading that is 20% pan wall is not slightly worse, it is a different number.

The fix is to get closer: at 100 mm a 12:1 instrument averages 8 mm, which fits inside anything. Note that the laser on many single-beam guns is offset from the optical axis, so at close range the circle is not centred on the dot.

Immersion depth, and where the error is really hiding

Liquid-in-glass thermometers are made for a specified immersion, and the specification is part of the instrument. Total immersion means the whole liquid column must be in the medium. Partial immersion means an etched line, commonly 50 to 60 mm above the bulb on the confectionery and dairy patterns sold to makers and 76 mm on many ASTM types, and the reading is only correct when the stem is submerged to that line. Read it shallower and part of the column sits in cool air, contracting more than the calibration assumed. The standard correction:

Correction = K multiplied by n multiplied by (bath temperature minus mean stem temperature), where n is the number of scale degrees of emergent column, K is about 0.00016 per degree for mercury in borosilicate and about 0.001 per degree for the organic, spirit-filled liquids used in modern non-mercury thermometers.

Work an example at balm temperatures. A reading of 70 C with 25 scale degrees emergent and a mean stem temperature of 35 C gives a correction of 0.14 C for mercury and 0.88 C for a spirit fill. That is the honest size of emergent stem error here: under a degree, and only worth caring about with a spirit thermometer read badly short.

Which is the point, because the real error is not in the column at all. It is conduction along the stem. A sensor 10 mm into a melt has a metal path running out to a 20 C room, draining heat faster than the shallow immersion replaces it. The process measurement rule of thumb is to immerse the sensor plus 10 to 15 probe diameters, so a 3 mm stainless probe needs 30 to 45 mm under. And on a general purpose probe the sensor sits in the last 10 to 15 mm of the tip rather than at the very point, so a probe dipped 15 mm may have its sensor out of the melt entirely.

Response time: what a probe tells you about a rising melt

A temperature sensor is a first order system. Manufacturers quote a time constant, the time to cover 63% of a step change, and for a 3 mm stainless probe in stirred liquid it is 4 to 10 seconds. Fine thermocouples manage under a second, liquid-in-glass takes 30 to 60. Getting within 5% of the true value takes roughly three time constants, so 12 to 30 seconds for a kitchen probe. On a rising melt that lag becomes a standing error, and to a good approximation the reading trails the truth by the time constant multiplied by the rate of change.

In a water bath the lag is negligible. On direct heat it is a real error in the wrong direction: the display says 80 C while the fat is at 83, you keep heating a few seconds more, and the melt goes past 85 with nothing on the bench telling you. That is a concrete reason a water bath is not merely gentler but easier to measure.

Two further points. A probe in still liquid has a much longer effective time constant than its specification, because the specification assumes moving fluid and a still boundary layer insulates the tip, so stir while you read. And a probe taken from the air and dipped in for two seconds reads neither the air nor the melt but a point on the curve between them, which is the commonest way two sound instruments end up disagreeing.

Three degrees, five degrees, ten degrees

The size of the gap tells you where to look, and looking in the wrong place is why people replace sound instruments.

Three degrees. Inside the combined specification of two ordinary instruments. A probe specified at plus or minus 1 C against a gun specified at plus or minus 2 C, or 2% of reading, is permitted a three degree gap before anything is wrong. Ice-point both. If one fails you have your answer, and if both pass you are comparing a surface with a bulk and the gap is physics, not error.

Five degrees. Too large for tolerance stacking, too small for a gross fault. Two causes account for nearly all of it: either the gun is reading an open stirred surface against a bulk probe, which routinely gives three to eight degrees, or the probe is shallow and being cooled along its stem. Fix the geometry first. Immerse to 40 mm, stir, wait fifteen seconds, read again.

Ten degrees or more. Not a calibration difference at all. Something structural is wrong, and the list is short: the infrared spot is catching bare metal or a steam plume, the probe is resting against the wall or base, the sensor is above the liquid line, the battery is flat, or the probe was cooked past its rating at some point and is permanently out. Check too that you are not reading the water bath rather than the melt.

The rule underneath all three: for any decision that governs a pour, trust an immersed, stirred, ice-point-checked probe and nothing else. That is what keeps a batch clear of grainy shea butter, tunnelling and dips and undissolved wax specks, all temperature faults in different clothes.

One reference, a quarterly check, an offset card

Nominate one instrument as your reference and protect it. The best reference in a small workshop is usually an ordinary digital probe used for checking and never for working: it does not sit in a hot jug for twenty minutes, does not get dropped, does not get washed a hundred times. If you sell what you make, an instrument with an accredited ISO/IEC 17025 calibration certificate costs less than most people expect and gives you traceability you can file.

Every three months, ice-point everything, then compare it all against the reference in a stirred oil bath at 70 C. Two points, because a zero check alone misses span error. A probe reading 0.0 C in ice and 72.5 C in a bath the reference calls 70.0 is 3.6% high on span, and will put every pour two and a half degrees hot.

Then write it down. An offset card is a strip of tape on the instrument or a page in the file carrying the instrument identifier, the date checked, the ice point reading, the reading at 70 C against the reference, the offset to apply and the date the next check is due. Note in each batch record which instrument was used. That habit turns a mysterious run of bad batches into one diagnosable cause, and it is the discipline that resolves batch to batch inconsistency.

What plus or minus 1 C buys, and what it does not

The tolerance comes from the job, not from metrology. The pour windows are seven to eight degrees wide, tins at 60 to 68 C and tubes at 65 to 72 C, and once you stay clear of both ends you cannot spend more than about a quarter of that window on measurement uncertainty. One degree either way is the practical answer, and the same figure appears in food regulation for the same reason. It is a sensible target for anyone filling lip balm tubes or working through a first lip balm.

What it does not buy is uniformity. A jug of melt is not one temperature: between wall and centre, and between surface and base, there can be several degrees, and a probe reads one point in that field. Stirring is what makes a single reading representative, so an unstirred melt measured by a perfect instrument is still a guess. Instrument accuracy is the smaller half of the problem.

Nor does it settle disagreements with a supplier's specification. Balm materials have melting ranges determined by a defined method rather than melting points, so a two degree gap between your thermometer and a carnauba wax data sheet is more likely a method difference than an instrument fault, as set out on melting point methods. Cooling has the same problem in reverse, in crystallisation and cooling rate. And none of it tells you whether your instrument is adequate for a regulatory file: what an assessor expects is traceable calibration, a document produced by a laboratory rather than a procedure you invent at the sink. The ice point check keeps batches consistent. It is not evidence.

The decision rule when two numbers refuse to agree: if only one instrument has been immersed in a stirred liquid and passed an ice check, there is no disagreement to resolve. There is one measurement and one guess.

Reviewed and updated 6 September 2026. Spotted an error? Tell us and we will fix and log it.