Scope. This page provides a controlled reader-run protocol and does not claim site-run measurements. For cooking, cooling, storage and reheating, follow the current official guidance linked on the page.
You did everything right. You put the probe in, you waited for the number, the number arrived, you took the food off the heat. Twenty minutes later you cut into it and it is a stage past where you wanted it, in a direction you cannot undo.
Nothing went wrong with your measurement. The problem is that the reading you took was correct at the moment you took it, and the food carried on cooking afterwards. Heat that was already inside the food when you removed it does not stop. It keeps moving inward, the centre keeps climbing, and it climbs for longer than most people expect.
The universal advice for this is "rest it", which is true, and comes with no number attached, which makes it almost impossible to act on.
Why the number kept rising after you took it out
Heat moves from hot to cold. While the food is on the heat, the outside is much hotter than the centre, and energy flows inward continuously. When you take the food away from the heat source, you remove the supply. You do not remove the gradient that already exists inside the food.
So the outer layers, which are considerably hotter than the centre, keep giving up their heat in two directions: outward to the room, and inward to the cold middle. The centre goes on rising until the whole piece has evened out enough that the inward flow stops, and only then does the entire thing start to cool.
Three consequences follow, and all three are worth holding on to.
The rise is not instantaneous. It happens over minutes, and the peak arrives some time after removal, not at removal. The gap between "off the heat" and "hottest in the middle" is the window in which people cut into things too early and conclude that resting does nothing.
The size of the rise depends on how much heat was stored. A large, dense piece cooked at a high surface temperature has an enormous reservoir of heat in its outer layers. A thin piece cooked gently has almost none.
A gentler cook produces a smaller carryover. This is the part that connects carryover to the rest of this site. If you cook something in a low oven or a water bath, the outside is only slightly hotter than the centre by the time it is done, so the gradient that drives carryover is small to begin with. High surface heat buys you speed and pays for it in an overshoot you have to predict. That trade is real, measurable, and almost never stated.
What to measure
The protocol is fixed before any run, and it is deliberately boring.
Variables. Mass, shape, and cut. Several masses of a comparable cut, so that mass is isolated. Several shapes at a comparable mass, so that surface-to-volume ratio is isolated. Then a set of genuinely different items: a thick fish fillet, a thin one, a lean roast, a dense braising cut, and a set custard, so the results span the range of things people actually cook.
Cooking method held constant within a series. Comparing carryover between two pieces cooked by different methods tells you about the methods, not about mass, so each series holds the method constant and varies one thing.
Probe. Calibrated immediately before and after each session, with both offsets logged. A fixed probe left in place through the whole run, logging continuously, rather than repeated insertions. Every insertion is a hole, and holes change what you are measuring.
Placement. Recorded precisely and identically within a series. More on this below, because it is the step most likely to invalidate everything.
Ambient. Room temperature, draughts, the surface the food rests on, and whether it is covered, tented or bare. All four change the outward loss and therefore the shape of the curve.
Duration. Logging continues well past the peak, so that the curve shows the rise, the peak, the plateau and the beginning of the fall.
Replicates. More than one run per condition, on separate days, with the spread recorded rather than averaged into a false smoothness.
The log format is set out in how we test.
The curves
This page publishes no curves of its own. What you are building is one temperature-against-time trace per condition, logged in Celsius with Fahrenheit alongside if you want both.
Worksheet fields: item and cut; mass; shape and thickness; cooking method and medium temperature; ambient temperature and resting surface; covered, tented or bare; probe placement; internal temperature at removal; peak temperature; time from removal to peak; temperature when logging stopped. Copy these labels into a notebook or spreadsheet and fill one row per run.
The thing we expect the curves to have in common is their shape: a rise that is steepest immediately after removal, flattening to a peak, followed by a long slow decline. What we expect to differ, sharply, is the height of the rise and the time to the peak, and we expect both to track mass and surface-to-volume ratio rather than anything else.
If the data contradicts that expectation, record the contradiction. That is the only reason to run the test.
Why one rule of thumb cannot work
The advice you will find elsewhere tends to take the form of a single figure: food rises by so many degrees, or rest for so many minutes. Both forms are broken in the same way, because both quietly assume a piece of food of a particular size and shape.
The governing quantity is the ratio of surface area to volume. A piece with lots of surface relative to its volume sheds heat outward quickly, so less of its stored heat goes into raising the centre and the whole thing equilibrates fast. A dense, compact piece has comparatively little surface, holds its heat, and drives more of it inward for longer.
This is why the same rule cannot serve a thin fillet and a large joint. It is also why shape matters independently of weight: two pieces of equal mass, one long and thin and one compact, will behave differently, and the difference is not a rounding error.
Three secondary factors also move the curve:
- Water content and density, which set how much energy the food stores per degree and how readily heat conducts through it.
- Whether it is covered. Tenting with foil slows the outward loss, which raises the peak and lengthens the plateau. Leaving it bare does the opposite. This is a lever you control.
- How aggressively it was cooked. As above: a steep internal gradient at removal means a large carryover, and a gentle cook means a small one.
A single number cannot encode any of that. Record a curve per condition and use it only to reason between comparable conditions; this page supplies the method rather than a universal carryover figure.
Probe placement, or the whole exercise is decorative
A badly placed probe invalidates carryover measurement more completely than almost any other error, because carryover is defined by the difference between the centre and everywhere else. Measure the wrong point and you are watching the wrong story.
The rules for placement:
- Aim for the thermal centre, which is the point that heats last. It is not always the geometric centre. In an irregular piece it sits where the flesh is thickest and furthest from any surface.
- Avoid bone. Bone conducts differently from muscle and a tip resting against it reads its own local story.
- Avoid fat seams and cavities. An air pocket reads dramatically low; a fat seam reads its own temperature.
- Approach from the side where possible, so that the probe shaft runs through the thickest part rather than diving straight down through a thin section.
- Watch out for shaft conduction. A metal probe shaft passing through a very hot outer layer conducts heat along itself to the tip, biasing the reading. Faster, thinner probes suffer less from this.
- Leave it in. For carryover specifically, insert once and leave the probe in place. Repeated insertions puncture the piece, release moisture, and change the object you are measuring partway through the measurement.
If your reading jumps around as you push the probe in and out, you are not seeing noise. You are seeing genuinely different temperatures at genuinely different depths, and the lowest reading you can find as you sweep through is usually the closest thing to the centre.
How to turn a target into a pull temperature
The arithmetic is trivial. Take the temperature you want the food to reach at its peak, subtract the expected rise for that piece under those conditions, and take the food off at the result.
The difficulty is entirely in the second term. Getting that estimate right requires knowing the mass, shape, cooking method, and ambient conditions, which is precisely why the protocol records a curve per case rather than printing a single number.
Two rules go with the arithmetic, and neither is optional.
Verify with the probe. Always. A projection is a prediction and needs an explicit accuracy range derived from the same logged runs. Use it only to decide when to start paying attention, not to declare the food done. If the projection and probe disagree, follow the probe and revise the projection.
An estimate is not a substitute for a thermometer. A figure worked out in advance cannot see your kitchen, your oven, the thickness of the piece in front of you or the draught from the window. It is an aid for setting your pull point before you start. The instrument is what tells you what actually happened.
The consequences of getting it wrong are asymmetric, which should influence how you use it. Undercooked food can go back on the heat. Overcooked food cannot come back. When you are uncertain, pull earlier and check, because only one of the two errors is recoverable. That asymmetry gets sharper the narrower the window, which is why it matters most in the 60–70°C band and hardly at all elsewhere.
Where carryover is small enough to ignore
Over-correcting is a real failure and it produces underdone food from people who have just learned about carryover and applied it to everything. Some situations genuinely have negligible carryover, and knowing which they are is as useful as knowing the curves.
- Thin, low-mass items. A thin fillet or a small piece of vegetable has very little stored heat and a great deal of surface. It equilibrates almost as fast as you can plate it.
- Anything cooked gently to its target and held there. If the food has been sitting in a water bath or a very low oven at the target temperature, there is no steep internal gradient, so there is nothing to drive a rise. This is one of the underrated advantages of low-temperature cooking, and it is a large part of why a gentle method has a wider practical window than a fast one.
- Food going straight into a cold environment. Something plunged into iced water is losing heat outward far faster than it is redistributing it inward. That is the entire mechanism of shocking.
- Braises and anything cooked well past a texture threshold. When the target is a state rather than a temperature, a small rise changes nothing.
Notice the pattern. Carryover is a problem created by a steep gradient, and steep gradients come from fast cooking. Cook gently and the problem largely dissolves, which is the argument this site keeps making from different directions.
The related question of what resting does, as opposed to what carryover does to the temperature, is a separate matter with its own mechanism and a good deal of folklore attached: what resting actually changes takes it apart.
In Plain English
Carryover is the continued rise in the middle of a piece of food after you take it off the heat, driven by heat already stored in its outer layers.
It matters most to anyone cooking something with a narrow window and a high price: fish, tender cuts, custards. It matters least on thin items and on anything cooked slowly to its target.
Use it by setting a pull temperature rather than a target temperature. Decide where you want the food to end up, subtract the expected rise for that particular piece in those particular conditions, take it off there, and keep the probe in to watch what actually happens.
What to do next: the next time you cook something worth caring about, leave the probe in after you take it off the heat and just watch the number for ten minutes. One observation of your own food, in your own kitchen, will teach you more than any published curve.
Where to go from here
Build your own carryover record, which is the only version of this that will ever be exactly right for your kitchen. For each dish worth repeating, write down the mass, the shape, the method, the ambient temperature, the internal temperature at the moment you took it off the heat, and the peak the probe reached while it rested. The difference between those last two is the carryover for that piece under those conditions. Cook the same dish a second time and you learn how repeatable it is. Three entries on a card will set your pull point more reliably than any rule of thumb, because they came off your own equipment.
How per-ingredient targets and windows are labelled, and what each provenance label means, is explained in how to read the Temperature Ladder.