The Deficit Loop: Why Your Calorie Target Should Change Every Week

A 10-minute read on why a fixed calorie target is a guess with an expiration date, and what it takes to replace it with a number that corrects itself.

---

TL;DR

Every calorie calculator on the internet — ours included — does the same thing: it takes your height, weight, age, sex, and a rough activity level, runs a formula, hands you a number, and walks away. That number is a population average applied to one person. It is wrong for almost everyone, by an amount nobody can know in advance, and it gets wronger as you lose weight.

The fix isn't a better formula. It's a loop:

1. You log your meals and step on a scale.
2. Each week, we compare what you ate against what your body actually did.
3. That comparison produces a measured number for your energy needs — yours, not a textbook's.
4. Your target moves to match, inside safety limits that we will not cross.

We call it the deficit loop. It has been running on real users since late July 2026. Some things it does that surprised even us:

• 41% of the target changes it has made were increases, not cuts. A closed loop is not a diet app that keeps taking food away.
• A safety floor blocked a proposed decrease 78 times. When the math wanted to go lower than is safe for a given person, the floor won. Every time.
• The median gap between what the textbook formula predicted and what a user's own logs implied was 564 kcal/day — a number large enough to be the entire difference between losing weight and not.

That last number is also where the hard engineering problem lives, and I'll be honest about it below: a gap that big usually has two possible explanations, and getting the wrong one is how weight-loss apps hurt people.

---

The problem: a static target is a guess that expires

Here is the standard experience. You use a TDEE calculator. It says 2,100. Subtract 500, you get a 1,600-calorie target, and off you go.

Three things are wrong with that number before you've eaten a single meal.

It was never accurate to start with. We took this apart in detail before: run the same person through eight calculators and you get answers 600–900 kcal apart, mostly because a five-item "activity level" dropdown is asked to encode everything from your commute to your fidgeting. Two people with identical height, weight, age, and sex can differ by several hundred calories a day, and no formula can see that difference.

It decays as you succeed. Lose 15 pounds and you are a smaller person who burns less at rest and less in motion. The target that produced those 15 pounds will not produce the next 15. This is the mechanism behind most weight-loss plateaus — not a broken metabolism, just an expired number.

Nobody tells you when it expires. The scale stalls. You assume you're doing it wrong. You aren't; your target is out of date. But the calculator that gave it to you has no idea you exist anymore.

A number that is wrong from day one, drifts further wrong every week, and never announces it — that is not a plan. It's an opening guess.

Closed loop, not open loop

There's a word for a system that acts and then checks whether the action worked: closed loop. Your home thermostat is one. It doesn't compute how much heat your house should need from square footage and outdoor temperature and then run the furnace for exactly that long. It measures the room, and it keeps measuring.

Almost every calorie tool is open loop. It computes an estimate and never looks at the result.

The deficit loop closes it. The measuring instrument is the thing you're already doing — your meal log and your weigh-ins. Put them side by side over a stretch of days and they tell you something no formula can:

You ate roughly this much. Your weight moved roughly this much. Therefore your body is burning roughly this much.

That is a measurement, and it beats an estimate. It quietly includes your genetics, your medications, your job, your fidgeting, your sleep, your GLP-1 dose, and the fact that you take the stairs — none of which any calculator will ever ask you about.

Do that every week and the target stops being a guess you're stuck with. It becomes a reading that gets sharper the longer you stay.

The catch: the same gap has two causes

Here's the part that makes this genuinely hard, and the reason most products don't do it.

Suppose the formula says you burn 2,200 kcal/day. Your logs say you ate 1,400/day for three weeks. Your weight didn't move.

Something is clearly wrong — the numbers don't add up. But there are two completely different explanations, and they look identical in the data:

1. Your real energy expenditure is far below the formula's guess. In which case the honest response is to adjust your target down, carefully.
2. You ate more than you logged. Weekend meals, the handful of chips, the oil the restaurant cooked in, the three days you didn't log at all. In which case your expenditure may be fine and the record is the thing that's broken.

Now consider what happens if a system assumes #1 when the truth is #2. It cuts an already-low target. The user is now told to eat even less, based on a measurement error. They get hungrier, adherence gets worse, logging gets worse, the next reading is even more distorted — and the system cuts again. That is a spiral, and it is dangerous. It is also, structurally, exactly what a naive "just close the loop on calories" implementation does.

So the loop's first job is not to move your target. It's to work out which question it's even looking at — and, when the evidence points at the record rather than the metabolism, to leave the target alone and go help you fix the record instead. That triage step is the core of the system and the thing we're keeping under the hood for now. But the user-facing consequence is simple and worth stating plainly:

We fix the measurement before we touch the food.

If your data is too thin or too contradictory to support a conclusion, the loop's correct output is no change. A target that holds still and a coach that asks a better question this week is a far better outcome than a confident wrong cut.

The guardrails, and what they actually cost us

A closed loop that can lower your calories needs limits that are not negotiable. Ours, and what they've done in production so far:

A hard floor. Every user has a minimum below which we will not set a target, derived from their body and their situation — never a flat 1,200 for everyone. When the loop's math proposed going lower, the floor stopped it 78 times. The loop does not get a vote. If your target is sitting on your floor and you're still not losing, the answer has to come from somewhere other than eating less — activity, or a conversation about whether the goal is right — and the coach is instructed to say so rather than squeeze.

A cap on how aggressive the deficit can be. The loop is not allowed to chase a fast result. A rate that would be impressive on a chart and miserable in a life gets clamped.

A speed limit on change. Targets move in modest steps, not lurches. In production the median applied change has been 100 kcal — the size of an apple. You should never open your phone to find your allowance slashed by 400 calories because of one strange week.

One exception to the speed limit, on purpose: if a previous state left someone below their safe floor, correcting upward is not throttled. Moving toward safety happens immediately; moving toward restriction happens slowly. The asymmetry is deliberate.

A human still reviews each batch. Weekly recalculation is not yet fully automatic. Every week's proposed changes are reviewed before any of them reach a user. We'll automate it when we trust it more than we trust that review — not before.

What this looks like if you're the user

You don't see any of the above. There is no dashboard, no settings screen, no "recalibrate" button. NanoRhino is a text thread; that's the whole interface.

What you see is that some weeks, a message arrives that says your target has moved and why — in plain language, tied to what you did. Something like "your logs and weigh-ins over the last three weeks put your real burn a bit higher than we assumed, so I'm moving you from 1,491 to 1,549."

That one is a real example, and I want to sit on it for a second, because it's the moment that convinced me this was worth building. The user is 77 years old, diabetic, on metformin. The loop found she'd been undershooting and raised her target by 58 calories. Her instinct on reading it was to eat less anyway — which is what a lifetime of diet culture trains you to do. The coach held the line, explained why more food was the correct move, and they landed on a plan she was happy with.

A tool that only ever subtracts could not have had that conversation.

Most weeks, of course, nothing happens at all. That's the system working. If your data doesn't justify a change, you shouldn't get one.

The numbers, honestly

Small company, early product, so here is exactly what we can and can't claim.

What our users are doing. Across everyone with at least two recorded weigh-ins (50 people): 76% are net down. Among the 32 who've been weighing in over a span of four weeks or more, 75% are down, averaging 7.5 lb over about six weeks. Of the longest-running 14, ten are down an average of 9.2 lb. The largest single result is 17.9 lb in 46 days.

What the loop itself has done. 34 applied target changes so far. Median absolute change 100 kcal. 14 increases against 17 decreases — the loop gives food back nearly as often as it takes it. The safety floor bound 78 times; the deficit cap bound 139 times.

What we can't claim. This is a self-selected group — people who weigh themselves at least twice are, definitionally, people who are engaged. There's no control group, no randomization, and nothing here is a clinical trial. And the loop only went live at the end of July, so most of the weight above came off before it existed. It would be dishonest to hand it the credit. What I'll claim is narrower and, I think, more useful: the loop is now the only thing that sets a calorie target in our product, it has moved 34 targets, and it has refused to move many more.

Ask me again in six months, with a larger n and a real cohort comparison. If it doesn't hold up, I'd rather write that post than pretend.

What we're still bad at

• Weekly recalculation is manual. A person reviews each batch. That's a deliberate safety choice today and a bottleneck tomorrow.
• The loop needs data to work. Someone who logs three days out of fourteen and steps on a scale once a month gets a frozen target and a coach nudging for more input — correct behavior, but it means the people who'd benefit most from precision are often the ones we have the least precision for.
• Our biggest constraint isn't the math, it's activation. The hardest problem in this product is not estimating energy expenditure. It's the gap between signing up and building a habit sturdy enough to measure. A perfect controller with no input signal controls nothing.

The principle underneath all of it

Ten years of building AI products taught me one thing that transfers cleanly to health software: when a system can measure something, it should never assume it.

Your body is running an experiment on itself every single week. It publishes the results on your bathroom scale and in what you ate. Nearly every calorie tool ever built ignores that data completely, in favor of a formula fitted to strangers.

We think the formula's only job is to get you through week one. After that, the best information about you is you.

---

Try it

Your first target still comes from a formula — start with our free Accurate TDEE Calculator, no signup or email. What happens after that is the part this post is about: start messaging your coach on WhatsApp or SMS, log your meals, step on a scale a couple of times a week, and let the number correct itself.

Pricing matches the philosophy: $10 per pound you actually lose, capped at $500 — no subscription, and the same pound is never billed twice. If the loop doesn't work for you, it doesn't cost you. Full pricing details here.

---

Written by Xiaomeng Li, Co-founder & CTO of NanoRhino. I'm an AI engineer, not a clinician — I have no medical or dietetic credential, and nothing here is medical advice. Calorie targets, energy-balance dynamics, and safe minimum intakes are clinical territory; if you have a medical condition, are on medication such as a GLP-1, are pregnant, or have any history of disordered eating, talk to a doctor or a registered dietitian before changing what you eat. Our system enforces per-user minimum intake floors and will not set a target below them.

Built with care by Link Heart Limited in Houston, Texas.