You're three weeks into a plan. You've logged most days, you've stayed near your number, and the scale is barely moving. Or it's moving faster than you asked it to, and you're hungry all the time.
Either way, the thought arrives at about the same moment: the number is wrong. And the move most people make next is to change it hard and change it now. Five hundred calories off tomorrow, or give up on the whole thing.
The first half of that thought is often right. The second half is where it goes wrong.
Your target was a guess from the start
Every calorie target begins as an estimate. A formula takes your sex, age, height, weight and how active you say you are, and hands back what someone shaped like you probably burns. It's a good formula, and it's still describing an average person rather than you.
So a target isn't a fact about your body. It's a hypothesis, and your weigh-ins over the next few weeks are the experiment that tests it. Finding out it was a bit off isn't failure. It's the system working.
What that framing also tells you is that a correction is only as good as the experiment behind it. Change the number on bad evidence and you haven't corrected anything. You've just swapped one guess for another.
Four things that have to be true first
There's a real trend to read. One weigh-in is mostly water, which is its own post. What you need is two weeks of readings, enough of them on each side to average, and a gap between what's happening and what you planned that's bigger than the noise. If you weigh most mornings, that noise works out to about half a pound a week, and a gap smaller than that can't be told apart from water, however much it feels like a stall. Graham uses that half pound as his bar whatever your cadence. If you weigh less often, your real noise is wider than his bar, so a verdict that you're off pace deserves a little more patience before you believe it.
The log is good enough to blame the eating. If the trend is off pace, the obvious conclusion is that the target is wrong. But that only follows if you actually ate roughly what the log says. With several days missing from the fortnight, the gap might just as easily be meals that never got written down. In that case the fix is the logging, not the number, and changing the number will make things worse. Graham's bar is ten complete days out of the last fourteen, and what counts as complete is deliberately easy to clear.
The last change has had time to work. A new target needs about two weeks before the trend can say anything about it, for the same reasons the first one did. Change it again after five days and you've learned nothing about either number. You're reacting to the scale, not testing anything.
The change is the right size. This is the one people skip, and it matters most.
Small enough to test, big enough to matter
When the evidence is good, how far should the number move?
The miss itself suggests an answer. A pound is roughly 3,500 calories, so a trend running half a pound a week behind plan works out to about 250 calories a day. That's the arithmetic Graham starts from, and since he won't call a gap under half a pound a week at all, 250 calories is about the smallest ordinary step. Then he bounds it at both ends.
Never less than 200 calories. A smaller change is hard to see. It sits inside the ordinary slop of a food log, and two weeks later you can't tell whether it did anything.
At most 400 calories. This is the bound I'd defend hardest. A big miss, a whole pound a week off plan or more, tempts you to move the number a long way in one go. But a miss that size is more often a sign that something upstream is off — the starting estimate was further out than usual, or logging has been quietly slipping — than a sign that you need a very different target overnight. So the step is capped. If the miss was real, the next fortnight will show it and the next step will follow. You get there in two moves you can check, rather than one you can't.
It works in both directions. If you're losing faster than you asked to, the same arithmetic raises the target, under the same limits. A trend running ahead of plan is just as much a sign the guess was off, and the answer to that is more food, not congratulations.
And there's a floor. However the trend reads, the target won't be proposed below a minimum set for safety rather than by the arithmetic. That's the one thing allowed to cut a step short: if the target is already close to the floor, the change stops there, even when that makes it smaller than 200 calories. If the target is already at the floor, the honest answer is that it shouldn't go lower, and that gets said plainly.
What Graham does with this
Graham never changes your target by himself. The weight trend is read in code, every guardrail above is checked in code, and when all of them pass the code works out one specific new number.
The one place he brings it up unprompted is the end of his weekly review, the first thing you see when you open the app in a new week. (Ask him how the plan's going in the meantime and he'll mention it then too.) It's a question, with a line on why: your trend's been running behind the pace you asked for; want me to take you to 2,000 calories? Say yes and the change is made. The tool that applies it doesn't take a number at all. It re-reads what the code proposed, so Graham can't swap in a figure of his own when he makes the change. (You can still set your own target any time by naming it, and he'll use yours. What he can't do is choose one for you.) Say no and he drops it. He doesn't bargain, and he doesn't keep asking.
When a guardrail is what's stopping him, he says which one. Not enough complete days, and he'll ask for a fuller week of logs before reading anything into the trend. Too soon since the last change, and that change still needs its two weeks. At the floor, and the target shouldn't go lower.
The reason for all of that structure is the moment this post started with. Week three, the scale isn't cooperating, and the number feels wrong. It might be. But a target you change slowly, on evidence, in steps you can check, is the only kind of correction that tells you anything. A target you change every time the scale disappoints you was never being tested at all.
— Ryan