Your Calorie Target Is Wrong. Here's How We Fix It.

Every calorie calculator on the internet does the same thing. You enter your age, weight, height, and activity level. It multiplies some numbers together. It gives you a target. You follow it for six weeks and nothing happens the way it predicted.

That’s because the formula doesn’t know you. It knows the average 30-year-old male who is “moderately active.” You are not that person. Nobody is.

We built something different.

TL;DR: Most calorie targets come from a 1990 formula (Mifflin-St Jeor) that doesn’t know your body composition, training load, or metabolic adaptation rate. Omnio watches what you actually eat, how your body actually responds, and builds a Bayesian model of your personal caloric response curve — anchored to DEXA scans when available. The target updates weekly as the model learns, with progressive confidence disclosure so you always know how much to trust the number.

The Problem with TDEE Calculators

The Mifflin-St Jeor equation — the one behind virtually every online TDEE calculator — was published in 1990 (Mifflin et al., 1990). It predicts your basal metabolic rate from four variables: age, sex, height, and weight. Multiply by an activity factor (usually 1.55 for “moderate activity”), and you get a total daily energy expenditure.

It’s a fine starting point. As a final answer, it’s terrible.

The equation doesn’t account for your actual body composition. It doesn’t know your training volume or how much that training costs you in calories. It doesn’t know that your metabolism adapts when you cut for eight weeks straight. It doesn’t know that you absorb nutrients differently from the next person, that your NEAT drops when you’re in a deficit (Rosenbaum & Leibel, 2010), or that your body partitions surplus calories into lean mass at a rate that’s genuinely individual (Trexler et al., 2014).

Most importantly, it never updates. You could follow a TDEE calculator’s advice for a year, gain 10 kg of fat, and it would still cheerfully tell you the same number tomorrow (adjusted for the new weight, of course — how helpful).

We wanted a system that actually watches what happens and adjusts.

What Omnio Does Instead

Omnio already tracks two sides of the equation:

What you eat. Daily calories and macronutrients from meal logging — photo-based AI analysis, manual entry, or MyFitnessPal sync.
How your body responds. Weight from your smart scale (Withings, OpenScale, etc.), body fat percentage, lean mass. And periodically, DEXA scans — the gold standard for body composition.

The missing piece was connecting these two streams and learning what the relationship looks like for you specifically. That’s what we’ve just shipped.

Learning Your Caloric Response Curve

When you set a body composition goal in Omnio — “gain 0.25 kg per week” or “cut: moderate” or “reach 15% body fat” — the system starts building a personal model of how your body responds to different calorie intakes.

Here’s the core idea: between any two body composition measurements, we know your average daily intake (calories and macros) and your average training load (from your wearable). We also know the outcome — how your weight, body fat, FMI, and FFMI changed. Each of these windows becomes an observation.

Stack enough observations together and a pattern emerges. Your pattern. Not the average person’s pattern, not a formula’s prediction — the actual measured relationship between what you eat and what your body does with it.

The model is Bayesian, which means it starts with reasonable assumptions (a 500 kcal surplus typically produces about 0.5 kg/week of gain) and gradually shifts toward your personal data as it accumulates. Early on, the recommendations lean on population averages. After a few weeks, they lean on you.

Your Training Load Changes Your Target

Here’s something most nutrition apps completely ignore: your calorie needs aren’t static across the week. A heavy lifting day costs more than a rest day. A high-volume training block costs more than a deload.

Omnio’s model includes your training load — both strength volume and cardio load (TRIMP) — as input variables. It learns not just your baseline TDEE, but the marginal calorie cost of your training. When your sustained training load changes, your calorie target adjusts automatically.

During a deload week the target drops. During a heavy block it rises. The input is a 14-day rolling average of training load, not today’s session — so a single workout won’t move the number, but a sustained shift will. This is deliberate: wearable calorie estimates are noisy (Shcherbina 2017; Fuller 2020), the body compensates for exercise via reduced non-exercise activity (Pontzer’s constrained-TDEE model), and the only honest ground truth for TDEE is what actually happened to your body over a window. Refunding individual sessions in real time is how most apps get it wrong.

DEXA Anchoring

Smart scales are convenient. They give you a weight and a body fat estimate every morning. But let’s be honest — scale body fat readings are noisy. They swing with hydration, time of day, and whether you looked at a glass of water.

DEXA scans are the opposite: infrequent but accurate. When you import a DEXA scan into Omnio, the model treats it as a high-confidence anchor point. Between scans, it uses your scale data to track trends, but it knows the composition split (how much is lean, how much is fat) is most reliable at the DEXA points.

This means the model can tell you things like “62% of your weight gain is lean mass at your current protein intake” — but only when it has the data to back that claim up. When it doesn’t, it says so. For more on how wearable calorie estimates compare against lab measurements, see Which Wearable Is Most Accurate?.

Progressive Disclosure (We Don’t Guess)

We could have shipped a system that gives you a number on day one. We didn’t, because a number without evidence is just a guess with extra steps.

Instead, the model tells you what it knows and how confident it is:

Week one: “Keep tracking — building your profile.” No target. No pretending.

Week two: “Your TDEE appears to be around 2,450 kcal/day.” An early estimate based on intake vs weight trend. If you’ve set a goal, you’ll see a preliminary target with a “low confidence” label.

Week three: A real target with “medium confidence.” The model has enough data to make a recommendation it’s willing to stand behind, sort of.

Week four and beyond: “High confidence.” Full composition breakdown if you have body comp data. Plateau detection kicks in. Training-load adjustments activate.

The confidence score is computed from sample count, data recency, and prediction error. It’s not a vibe — it’s a number, and we show you what’s behind it.

Plateau Detection

Metabolic adaptation is real (Muller & Bosy-Westphal, 2013). Cut for long enough and your body fights back — NEAT decreases, hormones shift, the rate slows. Most apps don’t notice. You just stop losing weight and wonder what you’re doing wrong.

Omnio’s model compares its predicted rate of change against what actually happened, every week. When reality diverges from prediction for two or more consecutive weeks, it flags it:

“Your weight change has stalled — you may be adapting. Consider a diet break or adjusting intake by ~200 kcal.”

And if you have body comp data, it can distinguish a real plateau from a recomp:

“Weight stable but FFMI still rising — you may be recomping rather than stalled.”

After eight or more weeks of sustained deficit, the system proactively suggests a maintenance phase. Not because a rule says so — because your data says the current approach is losing effectiveness.

Three Ways to Set a Goal

Not everyone thinks about body composition the same way. We support three entry points that all feed the same model:

Rate-based: “Gain 0.25 kg per week.” Direct, precise. For people who know what they want.

Direction + intensity: “Lean bulk” or “Moderate cut.” For people who know the direction but not the number. We map these to sensible defaults (lean bulk = +0.15 kg/week, moderate cut = -0.5 kg/week).

Body composition target: “Reach 15% body fat” or “FFMI of 21.” The system estimates a sustainable rate from your historical response curve and works backward to a calorie target. Requires enough data to be meaningful — the model won’t guess.

All three produce the same thing: a daily calorie target that updates as the model learns.

Safety Rails

We’re not going to recommend 900 calories a day because the math says it would get you to your goal weight fastest. The model has hard floors: 1,200 kcal/day minimum for women, 1,500 for men. If your goal would require going below that, we cap the target and tell you the goal rate isn’t safely achievable.

On the surplus side, recommendations cap at TDEE + 1,000 kcal. Beyond that, you’re mostly gaining fat and the model’s lean partition estimates stop being useful.

If you stop logging for two weeks, the goal pauses automatically. It doesn’t delete — it picks up where you left off when you start tracking again. But the gap gets thrown out, because we can’t learn from data that doesn’t exist.

What This Actually Looks Like

You set your goal. You log your meals. You step on your scale. Over the first few weeks, the system watches and learns. By week three or four, your nutrition page shows something like:

Target: 2,650 kcal/day (for +0.25 kg/week) | Confidence: High

And that number actually means something, because it’s derived from what happened to your body when you ate these foods at this training load. Not from a formula. Not from a population average. From you.

Your calorie target was wrong. Now it learns.

Sources

Mifflin MD, St Jeor ST, et al. (1990). “A new predictive equation for resting energy expenditure in healthy individuals.” American Journal of Clinical Nutrition, 51(2), 241-247. DOI: 10.1093/ajcn/51.2.241
Rosenbaum M, Leibel RL. (2010). “Adaptive thermogenesis in humans.” International Journal of Obesity, 34(S1), S47-S55. DOI: 10.1038/ijo.2010.184
Trexler ET, Smith-Ryan AE, Norton LE. (2014). “Metabolic adaptation to weight loss: implications for the athlete.” Journal of the International Society of Sports Nutrition, 11(1), 7. DOI: 10.1186/1550-2783-11-7
Muller MJ, Bosy-Westphal A. (2013). “Adaptive thermogenesis with weight loss in humans.” Obesity, 21(2), 218-228. DOI: 10.1002/oby.20027