In one sentence
VT estimates what you burn in a day from the one relationship no device guesses, how your weight moved against what you logged eating, anchored by a physiological estimate from your lean mass and training load, and combined into one figure with a range; the range is the headline.
Why not the watch's number
A wrist device's calories-burned figure is a model's guess: when seven wrist devices were tested against laboratory calorimetry, their energy-expenditure errors ran from 27 to 93 percent and none came within 20 percent (Shcherbina 2017). An error that size swamps any deficit a plan would set, so the engine excludes device calories entirely.
Where the method comes from
It is the energy-balance arithmetic of the research literature, not VT's invention. Hall's group built the mathematical model of adult metabolism that turns a change in intake into a weight trajectory, and showed that the response is slow, with half-times of about a year, and larger in people with more body fat (Hall 2011). The same group then ran the arithmetic the other way: in 140 adults who restricted calories for two years in the CALERIE study, a model whose only inputs were demographics and repeated body weights recovered the change in intake to within 40 kcal a day of the doubly-labeled-water and DEXA reference on average, with an individual root-mean-square deviation of 215 kcal a day (Sanghvi 2015). What is VT's is the way the pieces are put together for one person with sparse, imperfect data, and the care taken to say how sure the number is.
How it works, in plain words
Four parts. First, the measurement: across a window of days, your expenditure is your average logged intake minus the energy in the weight you gained or lost, spread over the days. The weight change is a line fitted through the whole window's mornings, never two mornings subtracted, and the energy per kilogram is not the folklore 7,700 kcal: early loss is water, glycogen and lean tissue and later loss is mostly fat, so the engine adapts the figure to the tissue being gained or lost, following Forbes's partition of lean and fat mass and Hall's derivation from it. The 3,500-kcal-per-pound rule that most calculators still use over-states weight loss badly against seven supervised experiments (Thomas 2013). Second, the window: a measurement like this is valid only within one regime, so the engine runs change-point detection on your raw intake and weight series and starts the window at the last committed change, a diet started, a training block begun, an illness, rather than pooling across it. Third, the prior: resting metabolic rate times an activity multiplier. Resting metabolic rate comes from a lean-mass equation (Cunningham's, the one popularly called Katch–McArdle) when your scale's lean mass has been corrected against a DEXA scan, and otherwise from an equation that needs only height, weight, age and sex (Müller or Mifflin–St Jeor); the engine never assumes you own a body-composition device. The activity multiplier maps your session-RPE training load through a curve with a ceiling, following the constrained-expenditure finding that total expenditure rises with activity at low levels and plateaus at high ones (Pontzer 2016, 332 adults in five populations measured with doubly labeled water). Fourth, the combination: the measurement and the prior are weighted by how sure each is and combined; a sparse or confounded window leans toward the prior, a dense clean one toward the measurement, and the output is a central figure with a credible interval.
What it takes into account
Confounded days are handled by mechanism, not by a generic outlier rule: a day marked unwell, an evening's alcohol, a poor night and a late meal each down-weight the day; menstrual water retention is a recurring shift and is offset rather than down-weighted; correlated causes are counted once. On a day you mark as a diet break, the engine counts what you ate but sets aside that morning's weight, which a refeed inflates with water, and the day cannot start a new window. When your weight is statistically steady, the engine takes a fast path, expenditure equals your mean reported intake, and the ratio of that to the prior becomes a logging-consistency signal: a ratio well below one suggests under-reporting, a ratio above one suggests the prior under-reads you, and the ratio cannot tell those apart on its own, which is why it is shown as a diagnostic and never as a verdict. A metabolic-adaptation arm reads the overnight heart rate, HRV and temperature against your baseline during a deficit; it ships observational, logging its signal without moving the estimate, because on eighteen months of the founder's data it could not be separated from the energy-balance measurement, which already captures adaptation as a slower weight change.
What it needs
29 of the last 36 days with both a weight and a logged intake. Without body composition, the prior falls back to the anthropometric equation; without a training log, the activity term sits at its floor; nothing is penalized for being missing, and a person with a scale and a food log is a first-class user.
What you see
The range first, a credible interval around a central figure. A confidence band built from named parts, data sufficiency, logging consistency, confounder load, estimate stability and prior basis, rather than one opaque score. The consistency diagnostic when the fast path is active. A wider interval, or no estimate, when the data is not good enough, rather than a confident wrong one. The Weight Loss Plan sets its deficit from the estimate; what you eat and how you train are separate decisions layered on top.
How far to trust the number
It is an estimate with an interval, not a measurement, and the interval is the honest part. The dominant error in practice is the food log. The engine has so far run on one person's eighteen months of records with no doubly-labeled-water reference, which establishes feasibility, not generalizable accuracy; a simulation study with known ground truth is in preparation and a cohort validation will follow. Several parameters were set by hand on a single subject and will be re-estimated as data accrues, under the method note's versioning rule.
Sources
- Vitality Tracker. How Vitality Tracker estimates your energy expenditure, method note v1.0 for TDEE Engine v2, 8 October 2026. doi:10.5281/zenodo.23224326
- Hall KD, Sacks G, Chandramohan D, et al. Quantification of the effect of energy imbalance on bodyweight. Lancet 2011;378:826–37. doi:10.1016/S0140-6736(11)60812-X60812-X>)
- Sanghvi A, Redman LM, Martin CK, Ravussin E, Hall KD. Validation of an inexpensive and accurate mathematical method to measure long-term changes in free-living energy intake. Am J Clin Nutr 2015;102:353–8. doi:10.3945/ajcn.115.111070
- Thomas DM, et al. Can a weight loss of one pound a week be achieved with a 3500-kcal deficit? Int J Obes 2013;37:1611–3. doi:10.1038/ijo.2013.51
- Pontzer H, et al. Constrained total energy expenditure and metabolic adaptation to physical activity in adult humans. Curr Biol 2016;26:410–7. doi:10.1016/j.cub.2015.12.046
- Shcherbina A, et al. J Pers Med 2017;7:3. doi:10.3390/jpm7020003. The equations (Cunningham 1980 and 1991, Müller 2004, Mifflin–St Jeor 1990, Forbes 1987) and the adaptation literature (Leibel 1995, Fothergill 2016) are cited in the method note.