The three TDEE numbers on your dashboard

Open your dashboard and you will find TDEE stated more than once, in different places, with different numbers. That is not a rounding error and it is not one of them being broken. They are three different measurements answering three different questions, and the moment you know which is which, the disagreement between them turns into one of the most useful things on the page.

This guide is the "how it is made" half. The dashboard itself tells you what to do with each number — the foldable How to read this box under each zone. Here we open the hood.

First: the dashboard is split by tense

The page has two zones, each with its own heading:

| Zone | Question | What lives there | |---|---|---| | Where you are right now | What do I burn today? | The dial, its four figures, the signal strip | | Where you have been | What did this stretch cost? | The two period panels and their charts |

Everything below follows from that split. The top zone is a snapshot. The bottom zone is a history. A snapshot and a history of the same thing are allowed to differ, and usually do.

1. TDEE right now — the dial

The big ember number with the needle under it. This is the app's answer to "put a gun to my head, what do I burn today?"

It is not an average of anything. Every other TDEE figure in the app takes a stretch of days and averages it. This one estimates a single day — the most recent one you have data for.

How it is made

Behind it runs a filter that walks through your whole logged history one day at a time, carrying four things it does not know and keeps re-estimating:

1. Your true weight — what the scale would say without water noise
2. Water and retention — treated as its own moving part that fades day by day
3. Your TDEE
4. Which way your TDEE is heading — whether your burn is climbing or falling

The physics is the model itself. Each day, your true weight is expected to move by (what you ate − what you burned) ÷ 7,700 kilos. So when the scale does something the filter did not predict, the mismatch has to be explained — and the filter splits the explanation between "that was water" and "our TDEE estimate was off," using how each has behaved for you so far. Learn that over hundreds of days and you get a burn estimate that keeps up on its own.

Days where you logged a weight but no calories are not thrown away: the filter substitutes your recent average intake and widens its uncertainty for that day, which is the honest thing to do.

"Your last ~3 weeks" is a simplification

The card says the figures read your last three weeks. For the four boxes — intake, deficit, hourly burn, expected loss — that is literally true: they average the last 21 days, anchored to your last entry, not to today, so a few missed days do not quietly shrink the window.

For the burn itself it is a simplification, and worth understanding because it explains most of the disagreement further down this page.

A filter like this has no hard edge. It does not include 21 days and exclude the 22nd; every day's influence fades gradually. We measured the real shape by nudging one day's weigh-in by a known amount and re-running:

| | | |---|---| | Your most recent weigh-in alone | 13% of the estimate | | Your last 21 days together | 44% | | Six weeks back | 64% | | Twelve weeks back | 93% | | Effective sample | about 33 days |

And there is a twist that matters. The influence has two halves with opposite signs, because a burn is read from a rate, and a rate needs two ends. A heavier-than-expected reading in the last ~19 days pushes the estimate down — you burned less than we thought. The identical reading two months ago pushes it up — you were heavier back then, so you have been losing faster since.

So the honest description is: your burn is anchored to the last three weeks and measured against a baseline reaching back a few months. Average speed and current speed, both inside one mechanism.

The shaded band

The wedge around the needle is one standard deviation either side — the room we honestly have to be wrong. It widens when your data is thin or noisy and narrows as it settles. In plain terms, expect the truth inside it roughly two times out of three.

We would rather that band be too wide than too narrow, and we checked: tested against simulated people whose true burn we knew, the engine's stated intervals aimed at 80 / 90 / 95% actually caught 88 / 93 / 95%. It errs generous. On the same test its typical miss was 88 kcal, against 101 for the best window-average method and 117 for the drift line on the TDEE chart.

What it needs before it appears

About three weeks of logged days. Below that, the block does not render at all — a headline nobody has earned yet is worse than no headline. The intake box needs at least 5 logged days in the window before it shows anything.

2. The period TDEE — All Time, a year, or one of your own periods

The blue and green panels in Where you have been. Each holds a period you choose: All Time, Year to Date, a completed year, one of your own diet periods, or Recent. Whatever you put in them, the TDEE is computed the same way.

How it is made

This is energy balance run backwards over a fixed stretch:

start weight  = average of the smoothed trend over the first few days
end weight    = average of the smoothed trend over the last few days
slope         = (end weight − start weight) ÷ days between those two midpoints
TDEE          = your average daily calories − slope × 7,700

A kilo of body weight is treated as 7,700 kcal (3,500 per lb). If you lost weight, the slope is negative and the subtraction adds it back — that is the energy your body spent but did not get from food.

Three details that are doing real work:

• The endpoints are averages, not single mornings. Up to seven days at each end, shrinking automatically for short periods. One morning's water weight standing in for the whole trend, at the two spots the smoother is least sure of, is how a number moves a hundred calories for no reason.
• The span is measured midpoint to midpoint of those two end windows — not first day to last day. Divide a midpoint-to-midpoint rise by an edge-to-edge distance and you flatten every slope you calculate.
• The trend it measures is the line you can see. There is one smoothed trend: the line on your Overview chart is the line this slope is read from — not a lookalike, and nothing you can tune. It is smoothed from both ends at once, forwards and backwards, so no single weigh-in gets to anchor it. (What that line actually is.)

What it is good at, and what it is not

It is the best answer available to "what did this whole stretch cost me?" It is the number for the record, the story, the screenshot.

It is a poor answer to "what should I eat today?", and the longer the period the worse it gets — for the simple reason that an average over six months includes a heavier body, a different season and a different you. Under six weeks the panel says so itself, with a caution line: still an estimate, improving as you log.

3. Recent — the same formula, a short rolling window

Set a panel to Recent and you get the identical calculation applied to a rolling window of the last X weeks, stepped with − and +. Default 6 weeks, adjustable from 2 to 52.

Recent never shows the reliability caution, because being short is the entire point of it — it exists to catch a change before it is certain. What it does have instead is two properties you should know:

• An average sits in the middle of its window. A three-week average is really "what you burned about ten days ago". While your metabolism is moving, ten days is enough to matter.
• It inherits every wobble of the scale. Half-kilo water shifts are ordinary, and multiplied by 7,700 that noise is large next to the signal. Measured: a 21-day window number moves about 42 kcal a day; the estimate on the dial moves about 24.

Six weeks is the steadiest read we could measure, which is why it is the default. Stepping through a few widths and watching which way the number leans is itself a useful move.

Why they disagree — and why the gap is information

Put the three side by side on a real record and you get something like: the estimate reading 3,419 while a 3-week Recent panel says 3,551, and a 6-week Recent panel says 3,402. Walk that forward across many days and the difference between the estimate and a 3-week window averages out small — a few dozen calories — while scattering several times that far in both directions. The gap is not a bias in one direction. It is three instruments with different characters.

Three causes, and it is usually a mix:

1. Where the number sits in time. The estimate sits at the edge — today. An average sits in the middle of its window. If your burn has been climbing, the average is behind you by half a window, by construction.
2. How much noise each one carries. A short window is jumpy for the reason above; the estimate holds water separately and moves about half as much.
3. They are answering different questions. All Time is a verdict on a journey. Recent is a read on the last month or so. The dial is a bet on today.

When the numbers sit close together, your metabolism has been steady. When they spread apart, something has changed — and the direction of the spread tells you which way. That is worth more than any of the three numbers alone.

The two triangles on the dial

This is the disagreement, drawn. The two small triangles on the dial's outer arc are your period panels' TDEEs, on the same scale as today's estimate — blue for Main, green for Secondary, the same colours as the panels below, so the marker you see is the number you can read.

Change the Recent window and the green triangle moves. That is deliberate: it is the one place in the app where you can watch "where I have been" slide against "where I am now" in real time. When a triangle sits far from the needle, your metabolism has moved since then.

They will sometimes overlap, and we leave them to. The case that causes it is Recent widening until it converges on All Time — two markers becoming one is the finding, not a collision.

What can move what

Worth stating plainly, because it is the thing most likely to trip you up:

• Nothing in "Where you have been" changes a number in "Where you are right now." Not the Recent stepper, not the period selectors, not the date range. The dial's window is its own.
• The one exception is cosmetic and intended: the Recent stepper moves the green triangle. It moves no number the top block states.
• Nothing you can click changes how the trend is drawn. There used to be a Smoothness slider. It was retired precisely so the line on the chart and the line behind the number could never drift into being two different things.
• The Main panel does drive the date range of the whole dashboard — every chart below shows the stretch Main is set to. Secondary changes nothing but its own column and its reference lines.

And the line on the TDEE over Time chart

Not a fourth answer — a picture of the first two. It takes the period formula and runs it over a rolling 28-day window for every day in your history, then draws a slow trend line through the result. It is there to show you the shape of your burn over months, which no single number can. Read the shape, not the individual points.

Which number for which job

| If you are… | Use | |---|---| | Deciding what to eat today, or setting a deficit | TDEE right now | | Checking whether something changed lately | Recent, plus the shape of the TDEE chart | | Describing the whole diet, or sharing a result | All Time or your own period | | Comparing two stretches against each other | Main and Secondary, side by side |

The mechanics are only half of it. Why an old TDEE quietly stops being true — and why more data does not always make it better — is the diet-side story: your TDEE has a shelf life.