How do I calculate my maintenance calories accurately?

Measure it rather than predicting it. Over a fourteen-day window, take the change in your smoothed trend weight, multiply by 3,500 to convert pounds to calories, divide by fourteen to get a daily energy balance, and subtract that from your average daily intake over the same window. The result is your actual maintenance, and it automatically includes every calorie your endurance training cost.

Why is my TDEE calculator wrong?

Because it predicts from body size and an activity dropdown, and neither captures what varies. Non-exercise movement is the most variable component of daily expenditure and no form can see it, prediction equations were fitted on general populations rather than athletes, and a single activity multiplier cannot distinguish a week with 30 miles of running from a week with none.

Does adaptive TDEE account for running and cycling?

Automatically and completely, without you logging a single session. It measures the outcome — what your weight actually did against what you actually ate — so every calorie the training cost is already in the number, whether or not your watch estimated it correctly and whether or not you remembered the session happened.

What if my weight goes up from water, not fat?

That is what the trend and the window length are for. Trend weight is an exponentially weighted average that damps single-day swings, and a fourteen-day window means a one-off water shift is diluted across two weeks. A large deliberate shift — starting creatine, refeeding after a diet — still distorts it, which is why the app labels those windows rather than silently absorbing them.

Adaptive TDEE: Measure Maintenance, Don't Predict It

Q: How long does it take to get an accurate maintenance figure?

Fourteen days, with at least ten of them logged and at least eight weigh-ins. Below that the weight signal is not separable from water and glycogen noise, and any number produced is a guess wearing a measurement's clothes.

Why the calculator you used is wrong

Every conventional TDEE calculator does the same two steps: predict resting metabolic rate from height, weight, age and sex, then multiply by an activity factor chosen from a dropdown.

Both steps carry error, and for a hybrid athlete the second one carries a specific and large error.

Prediction equations were fitted on general populations. Mifflin-St Jeor is the best-validated of them and still lands several hundred calories off for an individual. It tends to under-predict for people carrying above-average lean mass, which describes most people reading this.

The activity multiplier flattens exactly what varies most in your week. Consider one 180 lb athlete across two weeks. Week one: four lifting sessions, no running. Week two: the same four lifting sessions plus 30 miles of running. That running is roughly 2,300 kcal, or about 330 a day. Both weeks get "very active" and the same target.

If you are trying to run a +250 surplus, an error of 330 calories a day does not degrade the plan. It inverts it.

Non-exercise movement is the largest uncontrolled term. It is the most variable component of daily energy expenditure both between people and within one person across a week, and it adapts — eat less and you move less, without noticing. No dropdown can see it.

The formula

weightChangeLb = trendWeight(today) - trendWeight(14 days ago)
energyBalance  = weightChangeLb * 3500 / 14        // kcal/day, signed
maintenance    = avgDailyIntake - energyBalance

Worked example. Your trend weight went from 180.2 to 181.4 over a fortnight, and you averaged 3,100 kcal a day.

weightChangeLb = 181.4 - 180.2 = +1.2 lb
energyBalance  = 1.2 * 3500 / 14 = +300 kcal/day
maintenance    = 3100 - 300 = 2,800 kcal/day

You gained at 300 calories a day above maintenance. Your maintenance is 2,800. Not "about 2,800 according to a formula" — 2,800 as measured by your own body over two weeks.

The output is then smoothed with an exponentially weighted moving average at an alpha of around 0.2, so a single odd fortnight nudges the figure rather than yanking it.

The rules that make it trustworthy

Four constraints, and each exists because of a specific way the naive version goes wrong.

The window ends yesterday, not today. Today's food log is partial by definition. Including it drags the intake average down and reports maintenance a couple of hundred calories low every single morning — and worst at breakfast, which is exactly when someone checks.

Intake and weight are read over the same date range. Averaging intake over fourteen days and taking a weight delta over ten is a category error that produces a plausible, wrong number.

Minimum data: ten logged days and eight weigh-ins in the window. Below that the weight signal is not separable from water and glycogen noise. When the minimum is not met, the honest response is to fall back to a formula estimate and label it as estimated — not to compute the same arithmetic on four data points and present it identically.

Blank days are excluded and counted, never averaged as zero. A day you forgot to log is not a fasting day. Treating it as zero calories corrupts the intake average catastrophically and always in the same direction.

What it does that no other method does

It counts your training without you logging it. This is the property that matters for hybrid athletes and it is worth being explicit about why.

Wrist-based active energy carries wide error. Running calorie estimates depend on assumptions about your economy that are wrong for you specifically. Cycling without a power meter is guesswork. Every method that tries to add up your expenditure inherits every one of those errors and compounds them.

Adaptive maintenance does not add anything up. It observes the outcome. If you burned 4,100 calories on Saturday's long run, the scale knows, even if your watch said 2,900 and even if you never opened the app that day.

It absorbs adaptation. Metabolic rate drifts with sustained deficits and surpluses. A predicted number is wrong in a new way each month; a measured one re-derives itself every day.

It fails visibly. When there is not enough data it says so, rather than producing a number that looks identical to a well-supported one.

Where it goes wrong

Being honest about the failure modes is what makes the method usable.

Failure	What happens	What to do
Under-reported intake	Maintenance reads low, because the arithmetic attributes your unlogged food to your metabolism	This is the big one. The method is exactly as honest as your logging
Creatine started mid-window	Two to four pounds of intracellular water read as a surplus you did not eat	Discard the window. Restart the baseline after two weeks
A refeed out of a deficit	Three to six pounds of glycogen and gut refill, none of it fat	Expect it, label it, and do not act on it
Very large training-load change	The trend lags the new reality by a week or so	It self-corrects. Do not chase it mid-window
Under ten logged days	The number is not computed at all	Correct behaviour. A guess labelled as a measurement is worse than no number

Notice what is not on that list: your activity level being hard to classify, your metabolism being "different", or your watch being inaccurate. Those are the failure modes of prediction, and this method is immune to all three.

The number this makes possible

Once maintenance is measured, the surplus becomes meaningful — and for a hybrid athlete it needs to be, because the surplus is small.

A pound of lean tissue costs roughly 2,500 kcal. A trained intermediate adding 0.5 lb a week needs about 179 kcal a day of usable surplus. An advanced lifter at 0.25 lb a week needs 89.

Eighty-nine calories is smaller than the error bar on any prediction you could make. It is roughly a banana. You cannot steer at that resolution with a formula — and you do not have to, because you can steer by the trend instead. That is the argument for a +250 surplus rather than a +500 one, and it is why the check-in reports the surplus you actually averaged rather than the one you set.

Common mistakes

Recalculating every day and reacting. The window is fourteen days for a reason. Read it weekly.
Logging inconsistently and blaming the number. Ten days minimum, and they have to be honest days.
Eating back watch calories on top of a measured maintenance. The measured figure already contains them. Doing both counts the same run twice.
Chasing a maintenance figure that drifts up during a surplus. That is what is supposed to happen — more mass costs more energy. It means the plan is working.
Treating a two-week plateau as a stalled metabolism. Two weeks is one window. Three consecutive windows is a signal.

What to do next

Log honestly for fourteen days and weigh yourself most mornings. Do not change anything else during that fortnight — no new supplement, no diet change, no training block change. At the end you will have a maintenance figure that is genuinely yours, and every decision after it gets easier.

If you want the prediction as a starting point while you gather the fortnight, the hybrid TDEE calculator will give you one and tell you how much to trust it.