Marble computes its numbers in code from published equations and guidelines. Here are all 33 of them, each with what the app does and a link to the original source. The AI reads your plate and explains the numbers; it never sets them.
Marble is a tracking tool for healthy adults, not medical advice. If you're pregnant, managing a health condition, or have a history of disordered eating, talk to a doctor before following any calorie target.
- Your calorie target
- Rate of change and floors
- Protein, fat and fiber
- Carbs, alcohol and getting help
- Training
- Sleep
- Your weight trend
- Food figures
Your calorie target
Resting metabolic rate
Marble estimates the calories your body burns at rest with the Mifflin–St Jeor equation: 10 × weight (kg) + 6.25 × height (cm) − 5 × age, plus 5 for men or minus 161 for women. It is the equation with the smallest average error in validation studies, and it still misses by a few hundred calories for a meaningful share of people — which is why the app treats the result as a starting guess.
Activity multipliers
Resting rate is multiplied by a physical activity level from 1.2 (mostly sitting) to 1.9 (physically demanding work), the conventional bands derived from the FAO/WHO/UNU expert consultation on human energy requirements.
Reading that level from Apple Health
If you let Health set your activity level, Marble reads the last 14 days of workouts and active energy and picks the nearest of the same five levels. Active energy is what your body spent above rest, so the level it implies is (rest + active + the energy spent digesting the food that covers both) ÷ rest — digestion takes about 10% of what you eat, which the published multipliers already include. Days with a workout are read against the bands' own wording ("three to five days a week") as well, and the higher of the two readings wins; active energy only counts when Health has at least 12 of the 14 days, so a watch worn only for workouts can't make a fortnight look busier than it was. What reaches the formula is the level, never a calorie burn: your weigh-ins already contain every workout, so eating a burn figure back would count it twice.
Correcting the target from your weight trend
After enough weigh-ins, Marble compares the weight change your target predicted with the change your trend actually shows, and nudges the target. This rests on the energy-balance relationship between sustained intake and body weight described in the NIH dynamic model. The rule in one sentence: Marble moves the target only when the gap is clearly bigger than your own scale's day-to-day swings could explain (1.25 times, measured from your last eight weeks of weigh-ins), and then by half the gap, at most 250 calories, or at most 500 when the gap is twice what the swings could explain. Steps are at least two weeks apart and never go below the calorie floors under Rate of change and floors, and a step reaches the floor only on that same clear evidence. On a plan to lose, a stall doesn't wait for the next fortnight: Marble also reads a straight line through your weigh-ins since the target was set, every day, and lowers the target (half the gap, at most 250 calories) when you're losing under half the planned rate and the plan sits more than two standard errors from that line, or when the line has been flat for three weeks. Because that catches real stalls, the very first look waits for 1.5 times the scale's swings before cutting. These sizes are Marble's own cautious choices rather than published figures, checked against simulated people on the same NIH model.
When the scale stalls
Marble says where your weight stands against the plan (on plan, slower, stalled after three weeks with little change, or faster), read from the same four-week line, and never calls you on track when you're not. Weight that stalls is one of the strongest predictors of people giving up, and unmet expectations in the first weeks are another, so the app says it plainly and says what it will change. When days in the log have fewer than two meals, it suggests logging every meal for a week: days with at least two eating occasions recorded are the adherence marker used in photo-based food-logging research.
Before the target moves: a likely stall
From a week of weigh-ins, Marble fits a line through them with a term for water that rises and falls with the menstrual cycle (for women under 51, the median age at menopause), so the line's uncertainty includes what that water could hide. When it puts the chance that you're losing under half the plan at 80% or more, or the last one to two weeks look flat, Weight says what it sees, why water can hide a loss, the day the target comes down if the scale stays as it is, and one thing that's safe to do meanwhile. The target itself still moves only on the stall rule above. In Marble's test on thousands of simulated people (the Hall body model), people told a stall was likely were losing under half their plan 84 to 98% of the time.
Rate of change and floors
How fast to lose, and the deficit cap
Marble defaults to about 0.7% of bodyweight a week, allows at most 1%, and never sets a deficit deeper than 500 calories a day. Slower rates of loss preserve more lean mass and strength than faster ones.
Building muscle: a small surplus
The build-muscle goal sets calories 10% above your maintenance estimate. A small surplus with resistance training supports muscle gain while keeping the fat that comes with it low; larger surpluses mostly add fat. Your weigh-ins then correct the target, as for every goal.
Calorie floors
Targets never go below 1,200 calories a day for women or 1,500 for men, and deficits are held to 500 calories a day or less, following the NIH clinical guidelines on treating overweight and obesity in adults.
Who the targets are for
Marble's targets are built for adults at or above a healthy weight. Under 18, or below a BMI of 18.5 (the WHO and NIH line for underweight), the coach won't set a weight-loss goal or a faster pace and suggests a doctor instead: adolescents are better served by regular meals than by dieting or weight talk.
Calories per kilogram of body fat
To turn a weekly rate into a daily deficit, Marble uses roughly 7,700 calories per kilogram (3,500 per pound) of fat. This is the classic Wishnofsky figure; it overstates loss over long periods because the body adapts, which is one more reason the app corrects the target from measured weight rather than trusting the arithmetic.
Protein, fat and fiber
Protein targets
Marble sets protein at 1.8 to 2.2 grams per kilogram depending on the goal, counted on a reference weight (below) and never more than 35% of calories. A meta-analysis of 49 trials found benefits to muscle gain plateau around 1.6 g/kg, with the upper confidence bound near 2.2 g/kg; higher intakes, up to 2.3–3.1 g/kg of lean mass, are recommended when losing fat to protect muscle.
Protein for a higher body weight
Protein need follows lean mass, and body fat adds almost none of it, so grams per kilogram of total weight overshoot at a higher weight. Above a BMI of 25, Marble counts protein per kilogram of the weight at a BMI of 25 for your height (an adjusted body weight), the approach used in obesity and weight-loss protein research.
Protein for one meal
When the coach suggests a meal, it spreads the protein still to go today over the meals still to come, but never asks one meal to carry more than 60 g. Studies of how protein is used spread across the day support about 0.4–0.55 g per kilogram at each of four meals; past that, a bigger single serving adds little, and the rest of a gap waits for tomorrow.
Protein no more than 35% of calories
However the grams work out, protein never takes more than 35% of your calories, the top of the Acceptable Macronutrient Distribution Range (10–35%). That leaves room for carbohydrate and fat, which a low calorie target would otherwise squeeze out.
Dietary fat minimum and protein while dieting
Fat is set at about 25% of calories and never below 0.5 g/kg, and protein during fat loss follows the 2.3–3.1 g/kg-of-lean-mass range, both from the evidence review for natural bodybuilding preparation, the most demanding fat-loss setting studied.
Recomp: losing fat and building muscle at once
The recomp goal sets a slow loss of 0.25% of bodyweight a week (about 8–10% under maintenance) and protein at 2.2 g/kg. A review of recomposition studies found that people who already train can gain muscle while losing fat with high protein, progressive resistance training and a modest deficit. Because the scale barely moves when that happens, Marble shows your lift trend from Train beside your weight: estimated one-rep max (Epley), per lift.
Fiber goal
Today's fiber goal is 14 grams per 1,000 calories of your calorie target, the adequate intake set by the Dietary Guidelines for Americans. Most adults eat about half of it. Fiber is estimated from the photo along with the other numbers, so treat it as a rough guide.
Carbs, alcohol and getting help
Carbohydrate as a guide
Marble doesn't set a carb target. When the coach mentions one, it's the Acceptable Macronutrient Distribution Range: 45–65% of calories from carbohydrate, and at least 130 g a day, the amount the brain needs. It won't plan to a very-low-carb or other extreme diet.
Alcohol
Current US guidance is to drink less for better health; the previous edition set a limit of 2 drinks a day for men and 1 for women. The coach gives a drink's calories when asked, and helps fit up to 2 drinks into a day, keeping your meals and their protein. It never plans around more than that, never trades or skips a meal for drinks, and never cuts the next day to make up.
Where to get help
If something you say sounds like food has become hard (purging, bingeing, eating far too little), Marble stops coaching numbers and points you to people: the National Alliance for Eating Disorders helpline, 1-866-662-1235 (weekdays, 9am–7pm ET, licensed therapists), and 988, the Suicide & Crisis Lifeline, by call or text at any time.
National Alliance for Eating Disorders helpline; 988 Suicide & Crisis Lifeline (SAMHSA).
Training
A meal out: its share of the day
When you ask where to eat, Marble splits what's left of today's calories and protein across the meals still to come (three before 11 a.m., two before 4 p.m., then one), never going under 350 calories; the coach uses the same split for every meal still to come, so it never tells you to skip one. Before a workout the meal runs about 20% lighter; after one about 15% bigger. Protein for a meal around training is kept between 20 and 40 grams before and 30 to 50 after, in line with the ISSN's 20–40 g per meal. The coach is given this number and picks dishes near it; where a chain publishes its nutrition, those are the calories shown.
Estimated one-rep max
Train shows each lift's estimated one-rep max, worked out from your best completed set in each session with the Epley equation: weight × (1 + reps ÷ 30). Sets of more than 12 reps are left out, because the estimate drifts at high reps. It is an estimate for tracking progress, not a number to test.
Resistance-training energy estimates
Train shows active energy from a resistance workout as a range, based on the time logged and your body mass. The range uses the Adult Compendium's 3.5 MET value for varied 8–15-repetition resistance training and 6.0 METs for vigorous lifting. Resistance exercise has meaningful anaerobic work and its energy cost varies with rest, volume and technique, so Marble does not treat the range as precise or add it to your calorie target.
Sets per muscle each week
Your body counts each completed set once for every muscle the lift mainly works and as half a set for each muscle it assists, using the exercise library's muscle data, over the last 7 days. Its deepest colour starts at 10 sets a week, the amount a meta-analysis linked to the most muscle growth. It shows where your training went, not a score to hit.
Up next: 48 hours before the same muscles
Train puts a day up next only when none of the muscles it mainly works has been trained in the last 48 hours, the rest between sessions for the same muscle group that this position stand recommends. Among the days that are ready, it picks the one whose muscles have had the fewest sets this week. When none is ready, it suggests a rest day.
Sleep
How much sleep adults need
If you share sleep from Apple Health, the coach reads hours asleep night by night. It treats 7 hours or more as enough for adults, the joint recommendation of the American Academy of Sleep Medicine and the Sleep Research Society. Sleep never changes your calorie target.
Short sleep and hunger
After short nights people tend to eat more: a meta-analysis of sleep-restriction studies found about 385 more calories a day. That's why the coach calls a hungrier day after a short night expected rather than a lapse.
Short sleep and what you lose
In a controlled trial, dieters sleeping 5.5 hours lost about as much weight as at 8.5 hours, but about half as much of it was fat and more was lean mass. The coach mentions it to explain why sleep matters during fat loss, not to set a number.
Your weight trend
Why one morning's weight is noise
Body weight swings by a kilogram or more day to day from water, salt, glycogen and gut contents, so Marble smooths daily readings into a trend and shows that instead. Two-week changes in free-living adults are largely water and glycogen rather than fat.
Weekly weight rhythms
Weight tends to rise over weekends and fall during the week, which is why the app reads trends over weeks rather than reacting to any single reading.
Food figures
Restaurant and chain figures
When the coach gives a chain's dish its calories and macros, they come from the chain's own published nutrition (its website, app or nutrition guide), which US chains with 20 or more locations must provide. Marble keeps these figures with the page they came from and labels them "published" only when they came from that table.
Everyday foods
For plain foods (a banana, a cup of rice, a tablespoon of olive oil) the coach reads calories and macros per gram from USDA FoodData Central and multiplies by the portion. Photo estimates use the same table for their likely range.
The AI coach and meal estimates
Calorie and macro estimates from photos, and the coach's replies, are generated by an AI model (Anthropic's Claude) from your photo, your own log, and the methods above. They are estimates, not measurements. Barcode entries use the numbers printed on the package, via Open Food Facts. How accurate the photo reads are is measured on our research page.
Something here wrong or out of date? Write to support@getmarble.app.