Why Coffee Brews 15 Degrees Below the Kettle

This page contains affiliate links. If you buy through one we may earn a commission, at no cost to you.

One hundred degrees, and then falling at about two and a half kelvin a minute. But that is the easy half of the question and it is not where the temperature goes.

The water leaves the kettle at whatever the kettle says. It then passes through a cold dripper, a cold filter and 20 grams of cold coffee on its way into a cold glass server, and by the time it is sitting in the bed it is 15.55 kelvin below the figure on the display. A kettle that controls temperature to the degree has had its entire precision spent fifteen times over before the first drop arrives.

That is the useful answer, and it also settles a second question people ask a lot, which is why a cup from a shop can taste underdeveloped. The thermal problem is a home problem. A cafe's bigger batch loses less.

The mixing arithmetic, which is just a weighted average

A Fellow Stagg EKG Pro electric gooseneck kettle publishes 0.9 liters, to-the-degree temperature control, a Guide Mode with presets, a built-in brew stopwatch, and the two targets this argument needs: "Set exact targets like 205 degrees F for coffee or 180 degrees F for green tea." Two hundred and five Fahrenheit is 96.1 degrees Celsius, and that is the number on the display.

Now the pour. Heat capacity is mass times specific heat, and when two bodies mix the result is the average weighted by those capacities. Three hundred and twenty grams of water at 4.186 kilojoules per kilogram per kelvin is 1.3395 kilojoules per kelvin.

On the other side of the ledger, at room temperature: a glass server of about 300 grams at 0.84 is 0.2520, a steel dripper and mesh filter of about 120 grams at 0.50 is 0.0600, and 20 grams of dry grounds at about 1.6 is 0.0320. Together, 0.3440 kilojoules per kelvin, which is 25.7 percent of the water's.

So the temperature in the bed is 1.3395 times 96.1 plus 0.3440 times 20, all divided by 1.6835, which is 80.55 degrees.

Fifteen and a half kelvin, gone, before extraction has started.

Rinse the same equipment with boiling water first so that it begins at 90 degrees instead of 20, and the same arithmetic gives 94.86 degrees, a loss of 1.24 kelvin. The rinse recovers 14.3 of the 15.5. There is no other single action in pour-over technique with that leverage.

Three numbers worth putting on one scale

The reason this matters is not the 15.5 on its own. It is the 15.5 against the two things people do try to control.

The dial resolves one kelvin, as published. A minute of standing around with the kettle open removes 2.42 kelvin, taking a loss coefficient of 2.0 watts per kelvin for a 0.9 liter stainless kettle 76 kelvin above a 20 degree room. And the brewer removes 15.55, all at once, at the instant of the pour.

The brewer is 15.5 times the resolution of the instrument bought specifically to control this variable, and it is the equivalent of six and a half minutes of dithering compressed into one second. The dial is not useless. It is simply downstream of a much larger term that nothing on any of these listings mentions.

What 15.5 kelvin does to the coffee

Dissolution rates follow an Arrhenius dependence, exponential in inverse absolute temperature, and our page on why steam cannot get hotter than 100 degrees sets out that pattern and the conventional way of handling it in food work. For coffee solubles the activation energy usually lands somewhere between 20 and 40 kilojoules per mole.

Run the ratio between 369.25 kelvin and 353.70. At 20 kilojoules per mole the hotter water extracts 1.33 times faster. At 30 it is 1.54 times. At 40 it is 1.77.

So call it one and a half. Which means a recipe written for three minutes of contact at 96 degrees needs 4.0 to 5.3 minutes at 80.6 to arrive in the same place, and nobody brewing to a timer gives it that. The cup is not wrong because the grind is wrong. It is short because the clock ran out at a temperature the recipe was not written for.

And the fix is not to turn the dial up. Pushing the kettle from 205 to 212 Fahrenheit recovers 3.9 kelvin of the 15.55, which is a quarter of it, and does so by boiling the water, which brings its own problems. Rinsing the brewer recovers 14.3 kelvin for the cost of thirty seconds and a kettle full of water you were boiling anyway.

The equipment that supplies the thermal mass

A MAVO pour over coffee maker set is the clearest case, because it publishes the whole cold side as a list: "a pour over kettle, a coffee dripper, a glass coffee server with a measurement scale, and a reusable stainless steel filter mesh". Food-grade stainless for the carafe and filter, glass and steel for the dripper and server, and a heat-resistant sleeve on the kettle.

Every item on that list is a term in the 0.3440. And not one of them has a published mass. Four components, materials named for all of them, and no figure anywhere that would let a buyer work out what the pour is going to lose. That gap runs across the category, which is why the masses above are assumptions and why the figcaption spells out what happens when they move: a 200 gram server gives a loss of 12.1 kelvin and a 400 gram one gives 18.8.

The set is also right about the part it does address. Its record notes that the reusable mesh "lets more coffee oils and a little fine sediment through than paper does", which is a filtration property and is correctly separated from everything else. Steel versus paper changes what gets through. It does not change the temperature, except insofar as a steel mesh is a few more grams on the cold side than a paper cone.

Our roundup of pour-over coffee makers for more brew control is the place to compare them, and the one question worth adding to that comparison is how heavy the thing is, because that is the variable this page is about and it is the one nobody publishes.

A kettle claim that checks out, and a conflation that does not

A Bonavita 1L digital variable temperature gooseneck kettle publishes 1200 watts, a range of 140 to 208 degrees Fahrenheit, which is 60 to 98 Celsius, six presets, a 60-minute hold, 304 stainless steel, and a speed claim: it "brings water to a boil in about five minutes".

That one is checkable and it survives. One liter from 20 to 100 degrees is 334.9 kilojoules. At 1200 watts that is 279 seconds, so 4.65 minutes at perfect efficiency. Five minutes therefore implies about 93 percent of the element reaching the water, which is exactly what an immersed element in a stainless body should manage. It is the only speed claim in this category that is published alongside its own wattage and comes out consistent, and it deserves to be said.

The same record then makes a mistake that our site has already taken apart in another aisle. One field says "1-degree temperature accuracy" and another says control "runs in 1-degree increments". Those are not the same statement. An increment is resolution, which is how finely the display divides the world, and accuracy is how close the reading is to the truth. Our page on whether a 1 gram display means a 1 gram accurate scale works through that distinction in detail, and everything it says about a scale applies here unchanged. A kettle that steps in single degrees has told you about its display, not about its sensor.

Which is worth knowing precisely because of the 15.55 above. If the sensor is two kelvin out, that is 13 percent of the brewer's error and it is invisible. The dial's problem is not that it is imprecise. It is that it is measuring the wrong end of the process.

A ceiling that does not exist at altitude

The Bonavita's top setting is 208 Fahrenheit, which is 97.8 degrees Celsius, and the listing publishes no altitude setting.

At 1,609 meters, which is 5,280 feet, water boils at about 94.4 degrees Celsius, so 202 Fahrenheit. The top six degrees Fahrenheit of that kettle's published range cannot be reached there by any means, because the water leaves as steam instead. A kettle set to 208 in Denver will heat until it boils, sit at 202, and never satisfy its own controller.

The Stagg handles this, and it is the only product in the category that does. Among the settings its record lists are "boil scheduling, hold mode, altitude, and chime". An altitude setting is the correct response to exactly this problem: it tells the controller where the boiling point is so the dial and the physics agree.

There is one oddity in the Stagg's own write-up worth flagging, because it denies the existence of a part the feature requires. Its summary says it "lacks a built-in thermometer (not needed due to digital control)", and its drawbacks repeat "No built-in thermometer (but digital display shows temp)". Digital control of a temperature is a feedback loop around a temperature sensor. A kettle cannot hold 205 Fahrenheit to the degree without measuring it. What the listing presumably means is that there is no dial thermometer in the lid, which is a reasonable thing to say and not what it says.

The one feature that acts on the right variable

Of the nine kettles here, exactly one publishes a property that changes the cooling rate rather than the heating rate. A Secura stainless steel double wall electric kettle publishes 1.0 liter, 1000 watts, a fully stainless interior and "Double-wall construction keeps exterior cool and retains heat".

Retains heat is the operative half. A double wall roughly halves the loss coefficient, so on the same arithmetic as above the drift falls from about 2.42 kelvin a minute to about 1.21. In the currency of this page that is 6.4 minutes of grace instead of 3.2, which is useful if you brew two cups in sequence and irrelevant to the 15.55 if you do not rinse the dripper.

A Breville BKE820XL IQ kettle approaches the same problem electrically instead: 7 cups, which is 57 ounces or 1.686 liters, five presets including one for French press, and a keep-warm function that "maintains the water temperature for 20 minutes". Twenty minutes against the Bonavita's sixty, and its own drawbacks name that as the limitation.

Nobody publishes what a hold draws, which is a pity because it is a small number and an interesting one. Holding a liter at 96.1 degrees against a loss coefficient of 2.0 watts per kelvin takes about 152 watts continuously, which is 12.7 percent of a 1200 watt element. A hold is not the kettle reheating. It is the kettle idling at an eighth of its power, and that is the figure that tells you whether an hour of hold is a reasonable thing to ask of it.

Two speed claims that do not survive their own arithmetic

Here is where the records get interesting, and both problems are the same shape: a boil time published without the wattage that would support it, or alongside one that will not.

An OVENTE 1.5L glass electric kettle publishes 1100 watts, 1.5 liters, borosilicate glass, a 360-degree base, and "Boils water in 7 minutes or less".

One and a half liters from 20 to 100 degrees is 502.3 kilojoules. Seven minutes is 420 seconds, so the claim requires 1,196 watts delivered into the water. That is 8.7 percent above its own published element, at perfect efficiency, with nothing left over for the glass or the base. The claim is achievable at a fill of 1.379 liters, so it is true of 92 percent of the kettle and not of the kettle.

It also publishes "50% more efficient than stovetop kettles" with no burner named, and gas and electric hobs sit far enough apart that there is no arrangement of numbers that makes that checkable in either direction.

A Cosori stainless steel electric kettle is the harder case, and it is the only one of these where the published figure is not merely optimistic but unavailable. It publishes 1.5 liters, a seamless 304 stainless interior, a two-level lid, shutoff 30 seconds after boiling, and "Quick boil in 3-6 minutes for 1.5L". It publishes no wattage at all.

Work backwards. The same 502.3 kilojoules in three minutes needs 2,790 watts. Our page on whether kettle wattage really matters establishes the ceiling that blocks it: a standard North American outlet on a 15 amp circuit allows 1,800 watts at the absolute maximum. The fast end of this kettle's own published range needs 1.55 times the entire circuit.

Even at that ceiling, with 100 percent of it reaching the water, 1.5 liters takes 279 seconds, which is 4.65 minutes. So everything from 3 to 4.65 minutes of a published 3-to-6 range cannot happen on the supply the appliance is sold for, and the half of the range that can is the slow half. Our roundup of temperature control kettles is the place to compare the controllable ones, where a published wattage is the figure that makes a boil time mean anything.

Why a shop's cup can taste thin, which is a different fault

The obvious explanation for an underdeveloped cup from a cafe is that their water cooled on the way in, and the arithmetic says the opposite. The water's heat capacity grows with the dose and the brewer's does not, so a bigger brew loses proportionally less.

Take a 60 gram dose with 960 grams of water into a larger brewer, say a 450 gram server and 180 grams of steel. The water is now 4.0186 kilojoules per kelvin against a cold side of 0.564, and the result is 86.73 degrees. A loss of 9.37 kelvin instead of 15.55. A cafe batch starts 6.2 kelvin better off than your single cup does, and that is before anyone rinses anything.

So if the cup is thin, look at the ratio rather than the temperature, because strength and extraction are two separate numbers and the ratio moves one and not the other. Twenty grams of coffee taken to a 19.5 percent extraction yield has dissolved 3.90 grams of solids, however much water went through it. At one part coffee to fifteen parts water, 300 grams in, about 40 retained by the spent bed, the beverage weighs 260 grams and carries 1.500 percent dissolved solids. At one to seventeen, 340 in and 300 out, the same 3.90 grams is 1.300 percent.

Identical extraction, 13.3 percent weaker. A cup brewed that way is not underextracted at all. It is correctly extracted and thin, and those two faults taste different: thin is flat and watery, underextracted is sour and hollow. Our comparison of a coffee maker versus pour over covers the control side of this, and the useful addition is that the two faults have different fixes. Thin wants a tighter ratio. Sour wants a hotter bed or a longer contact, and the hotter bed is the 14.3 kelvin sitting in a rinse.

The honest limit

What is exact here is the mixing. Conserving heat between two bodies gives the average of their temperatures weighted by their heat capacities, so the 80.55 and the 94.86 follow from masses and specific heats with no modelling in between. The two boil-time arithmetics are exact too, because they are an energy divided by a published power, and the 2,790 watts and the 1,196 watts are what the claims require rather than what anything delivers.

What is representative, and it is the load-bearing assumption of the whole page, is every mass on the cold side. No listing in this category publishes the mass of a dripper, a server, a filter or a kettle, and those masses are the entire 15.55 kelvin. The figures used here are ordinary for glass and steel equipment of that size, and the sensitivity is worth repeating: a 200 gram server gives a loss of 12.1 kelvin and a 400 gram one gives 18.8. The direction and the order of magnitude are firm. The exact number is proportional to a guess, and it would stop being a guess the moment anyone printed a weight.

Also representative: the loss coefficient of 2.0 watts per kelvin behind the 2.42 kelvin a minute and the 152 watt hold, the specific heats of 0.84 for borosilicate and 1.6 for dry grounds, the 2 grams of water per gram of coffee retained by a spent bed, and the activation energy band of 20 to 40 kilojoules per mole. That last one is why the extraction ratio is given as 1.33 to 1.77 rather than as a single figure, and anyone quoting one number out of that band is overstating what is known.

What this page deliberately does not compute is a flavor outcome. Extraction yield is a mass fraction and taste is not, and the mapping between them depends on the bean, the roast and the water, none of which any product here publishes. Nor has it tried to compute the temperature profile during the brew, which would need a flow rate through the bed and a bed permeability, and the only product that publishes anything adjacent is a pour-over set describing its filter as metal. The 19.5 percent is a target that the industry uses, not a measurement of anything.

And the residue is three sentences. Rinse the dripper and the server with the water you already boiled, because it is worth 14.3 of the 15.5 kelvin and nothing else is close. Treat the dial as a sensible last adjustment rather than the main one, and remember a published increment is not a published accuracy. And if you live high up, check whether your kettle knows it, because exactly one of the nine here has somewhere to tell it. Our roundup of gooseneck kettles is where to compare the pouring hardware itself.