You were careful yesterday. You logged everything, you came in under your target, you went to bed quietly pleased with yourself. This morning the scale reads two pounds heavier than it did the day before.
The conclusion most people reach standing there is that something they did was wrong, or that their body is uniquely broken, or — most often, and most expensively — that none of this works and there's no point keeping it up.
I'd offer a duller explanation. You didn't measure the thing you think you measured.
One reading is mostly water
A bathroom scale is an honest instrument pointed at a very noisy quantity. What it weighs is you plus everything currently inside you: water, the salt that's holding onto it, what you've eaten and not yet digested, the glycogen in your muscles and the water that rides along with it. Day to day that moves a pound or two in either direction, and none of it is the thing you're trying to watch.
The arithmetic on the other side is much less forgiving. A pound of body fat runs about 3,500 calories — that's the figure Graham's own calculator uses — so two pounds of it appearing overnight would mean about seven thousand calories arriving on top of everything you burned. You would remember that. Whatever the scale is reporting this morning, it isn't that.
What a week of them buys you
If one reading carries roughly a pound of noise, the obvious move is to stop reading one.
Average a week of them and the noise falls by about the square root of how many you took. Seven readings, √7, call it two and a half — so a week's average carries something nearer four tenths of a pound of slop instead of a full pound. That's the entire reason a trend line exists and why it isn't simply yesterday's number drawn slightly smaller.
You can't stop there, though, because the thing you actually want isn't a weight, it's a rate — and a rate is the difference between two of those averages. Subtracting one noisy number from another re-inflates the noise, by about √2.
Run the whole chain: about a pound on a single reading, divided by √7 down to roughly four tenths, multiplied by √2 back up. You land at about half a pound per week of noise on a fortnight's measured rate — at best, since that square root assumes each morning is independent of the last, and mornings aren't quite. A salty Saturday is still there on Sunday.
Half a pound a week is the floor, if you weigh most mornings
That figure is a constant in the code, and it decides when Graham speaks. A deviation from plan smaller than half a pound a week is indistinguishable from water, so he calls the fortnight on pace and proposes nothing. He doesn't tell you you're slightly behind, because he has no way to know that you are.
Half a pound is the floor if you weigh most mornings. It isn't a universal constant, and the derivation above is what says so: the √7 was seven readings. Weigh twice a week instead and the same chain lands nearer a pound, so the honest floor for that person is twice as wide as the one the app applies — which is worth knowing if you weigh rarely and get told you're off pace.
What the floor isn't is personal. It falls out of how noisy one reading is and how many of them you took — measurement, not metabolism — so it isn't something your own history could tune. A longer window genuinely does resolve a finer rate, because noise on a slope falls with the span as well as the count. What no amount of watching you can do is make a fortnight of weigh-ins say something a fortnight of weigh-ins doesn't contain.
Two rules for rounding, because there are two jobs
One small thing inside this took a bug to find, and it turned out to have two answers rather than one.
The average that feeds the rate is kept unrounded until the very last moment. It's tempting to round each week to a tenth of a pound — a tenth is all you'd ever display — and then subtract the two. Do that and the two rounding errors can point in opposite directions, and you manufacture up to a tenth of a pound per week of trend out of nothing whatsoever. Against a tolerance of half a pound, a tenth is a fifth of the entire budget, invented by tidying up. So there the subtraction runs on the full numbers and the rounding happens once, at the end.
The weekly table on the plan tab does the opposite, deliberately. It prints each week's average, a target derived from the week before it, and the gap between them, side by side — and a reader who subtracts the first two columns has to land on the third. So those averages are rounded up front, and the gap is the difference of two already-rounded numbers.
Both rules are right for their job and they cannot be the same rule. The cost is real and worth saying out loud: the printed gap can carry a tenth of a pound that the rate engine wouldn't, because that tenth is the price of three columns that agree with each other. A number that is only ever subtracted should stay raw. A number sitting next to its own operands has to be rounded first.
What that looks like from the outside
Tell Graham you weighed 192.4 this morning and he logs it and says almost nothing back. Got it — 192.4.
That brevity is a rule, not a mood. He's barred from reading anything into a single number — no direction, no pace, nothing about it starting to take shape. Every claim about where things are heading comes from the trend calculation or it doesn't get made, and one morning is not enough to make one with.
The plan tab says the same thing in stages, and says it plainly rather than drawing a confident line through two dots. With nothing logged yet, the trend builds from there. With one reading: one reading so far — a few more and the trend line appears. After that: building the trend — a couple of weeks of weigh-ins makes it readable. The weekly table will show you a gap against target as soon as there's a week on each side of it to compare, because that's a description of two weeks and you're entitled to see it. What waits is the verdict — on pace, off pace, and any suggestion that you change your target — and that needs two full weeks with at least two readings on each side before Graham will say it out loud.
Read less, rather than weigh less
People are often told to stay off the scale, and the reason is a good one: for plenty of people a daily number is a daily verdict, and the morning goes wherever it points. If that's you, that advice is right and none of the arithmetic above changes it.
But notice what the math actually asks for, because it isn't daily weighing. It's more than one, averaged. Three mornings a week already gives you an average rather than a number, and Graham's own setup asks for a weekly weigh-in rather than a daily one. Nobody needs to step on a scale every day to have a readable trend — a week takes whatever landed in it.
So the rule isn't weigh less. It's read less. Whatever your cadence, let this morning be one of several things that go into the week rather than a report card on yesterday — the same way a single missed day isn't a report card on the fortnight.
This morning's two pounds were never the thing you were trying to measure.
— Ryan