Calories in, calories out is simple math with messy measurements

Calories in, calories out (CICO) is the observation that this relationship is an accounting identity. It has to balance, in the same way a bank balance has to equal deposits minus withdrawals. If more energy comes in than goes out, the surplus is stored. If less comes in than goes out, the shortfall is drawn from stored energy. There is no version of human metabolism where this fails to hold.

What CICO does not say is that hunger, food quality, sleep, hormones, medications, and behavior are irrelevant. They are not side issues, they are the mechanisms that act on both sides of the equation. Sleep and stress shift appetite and how much you end up eating. Food composition changes satiety and how easy a given intake is to hold for months rather than days. Medications and hormonal state can move resting energy needs meaningfully. None of that breaks the accounting, it explains why hitting a given number of calories in, or a given amount expended, is so much harder for some people and in some seasons of life than others.

The intake side feels like the easy half to know, but it carries real error. Packaged food labels are allowed a regulatory tolerance rather than reporting an exact figure, so two products with the same printed calorie count are not guaranteed to match to the calorie. Home cooking multiplies the uncertainty further: recipe yields, ingredient substitutions, cooking oil that never gets measured, and portion sizes eyeballed instead of weighed all add error before a single number is logged. Restaurant meals are worse still, since posted calorie counts are estimates from a test kitchen, not a measurement of the plate in front of you, and normal preparation can vary from that estimate by a wide margin.

Ordinary logging error stacks on top of all of that: a forgotten cooking oil, a rounded portion, a snack that never made it into the log. None of this makes logging pointless, it means a single day’s number deserves less confidence than it looks like it has. It is also why Wellness Project treats a day with no log entry as unknown rather than zero, the average only divides by days you actually logged, while a day you deliberately log as 0 calories, a declared fast, counts as a real zero. Conflating the two would quietly drag a maintenance estimate toward looking too low.

The expenditure side is built from several pieces, each with its own error. Resting energy needs (the calories spent just staying alive) make up the largest share for most people but are only estimated by a formula unless directly measured, which almost nobody does. Non-exercise activity, walking, standing, fidgeting, everyday movement, is a genuinely large and genuinely variable share of the total that no formula sees. Structured training adds its own chunk, and even the energy cost of digesting food itself (thermic effect of food) is a real, if smaller, contributor. On top of all of it, expenditure changes as body mass changes: a lighter body generally burns somewhat less at rest and during the same activity than a heavier one did.

A formula like Mifflin-St Jeor gives a reasonable starting estimate for the resting piece, see TDEE for how that calculation works, but it is a population average applied to an individual. A device-reported number, imperfect as it also is, gets you closer to your own expenditure than a formula alone ever can, which is what calories burned accuracy covers in detail.

Because both sides of the equation are estimates, and because scale weight itself is not a direct readout of stored energy, day-to-day weight almost never matches what the math predicts. Water retention from sodium intake, glycogen stores that shift with carbohydrate intake and training, whatever is currently in the gut, fluid shifts tied to the menstrual cycle where relevant, and post-training inflammation can all move the scale by more than a real day’s energy imbalance ever could.

None of that means the equation broke. It means the scale is measuring total body mass, water included, not body fat directly, and water moves on a much faster and noisier schedule than fat does. That is the entire reason Wellness Project reads a smoothed trend rather than comparing one morning’s weight to another’s: a single endpoint-to-endpoint difference is mostly noise, while a trend line averages enough of it away to show the real direction underneath.

The figure that roughly 3,500 kilocalories of net deficit or surplus corresponds to one pound of body fat is a widely used energy-conversion heuristic, not a precise physical constant. It comes from the approximate energy density of fat tissue, and it is useful for rough planning, but it does not mean one pound of scale movement is automatically one pound of body fat, since scale weight includes water and other mass that swings independently of fat. It also does not stay perfectly fixed through a long phase, because both body composition and expenditure change as weight changes, so the same 3,500-calorie gap does not convert to identical results at the start and the end of a multi-month cut.

Wellness Project’s adaptive estimator applies roughly 3,500 kcal per pound as a practical conversion factor, and only ever against a smoothed multi-week weight trend, never against a single day’s scale reading. Applied that way, to a trend rather than a data point, the approximation is reliable enough to be useful. Applied to one morning’s number, it is not.

That second line is the same accounting identity from the top of this page, rearranged and solved for the expenditure side instead of the outcome side. Since intake can be logged and the weight trend can be observed, the equation can be worked backward to estimate maintenance instead of forward to predict weight change.

A few worked illustrations, labeled as illustrations and not as targets for anyone to chase: if the smoothed trend is flat, estimated maintenance lands close to average logged intake. If someone is losing roughly 0.5 pounds a week while averaging 2,000 calories a day, the math implies their average expenditure sits somewhat above 2,000. If the trend is rising instead, maintenance is implied to sit below whatever intake is being logged. None of these numbers are a prescription for what to eat, they are a read on what already happened. For how big a deficit to actually run toward a goal, see what is a calorie deficit.

A single day’s number can look precise, calories logged to the exact digit, a device readout with a specific figure attached, without being accurate. Precision and accuracy are different things: a food label’s tolerance, a portion estimate, and a device’s own margin of error do not go away just because the final number on screen has no rounding in it.

A consistent multi-week record beats a precise-looking single day on both sides of the equation. More logged days average out portion and label error on the intake side. More weigh-ins and a longer device history average out water noise and formula error on the expenditure side. Consistency, not decimal places, is what actually improves the estimate.

Food logging, body-weight history, and connected device activity and burn data all feed the same estimate rather than living as separate numbers to reconcile by hand. Logged meals build the intake side, weigh-ins build the smoothed trend, and device data grounds the expenditure side against something closer to your own body than a formula. The result surfaces as your maintenance calories estimate, and the same three signals sit together on one timeline in the app’s Calories Burned view.

U.S. FDA, 21 CFR 101.9, Nutrition labeling of food (calorie declaration and compliance)

Urban et al., Accuracy of stated energy contents of restaurant foods, JAMA (2011)

Thomas et al., Effect of dietary adherence on the body weight plateau, American Journal of Clinical Nutrition (2014)