You're at someone else's table. They've made a casserole, you're on your second helping, and nobody weighed anything. Or it's a work lunch, or a burger you made without measuring the patty, or a handful of something out of a shared bowl.
Nearly every piece of tracking advice has the same answer for this, which is to weigh it. That advice is correct and it is useless here, because the food is already eaten and the scale was never in the room.
So you do the thing everyone does. You skip the meal, tell yourself you'll catch up tomorrow, and now the day has a hole in it. A day with a hole isn't a day you can learn anything from, which is the quiet start of not opening the app again.
I'd rather you logged it badly.
A good estimate says what it assumed
There's a version of guessing that's useful and a version that isn't, and the difference is whether the guess shows its work.
"Burger and fries, about 800 calories" is a number you can't do anything with. "About 800 calories, assuming a single patty and modest fries" is a number you can correct in four words. If it was a double, you say so, and the whole thing moves.
That's the rule Graham runs on. When he constructs a number from a description rather than reading it off a label, he has to say the reference he used. You told him "burger and some fries, homemade, no clue how big" and he'll come back with something like called it about 800 calories — single patty, modest fries. If it was a double, say so.
The assumption is the useful part. It turns a guess into something with a handle on it.
The error you're worried about isn't the one that gets you
I think most tracking advice has this backwards. The error people fret over is the random one — was that scoop of mashed potato a half cup or closer to three quarters. That error is real, and it's also about as likely to run high as low, and across a week of meals it doesn't do much to your trend.
The errors that actually move your numbers are the ones that lean the same way every single time.
Restaurant food is the obvious one. People consistently guess restaurant portions low, because the plate looks like a normal plate and the oil and butter are invisible. So Graham deliberately leans about 20% high on restaurant food. Not because he's pessimistic, but because an unbiased guess about restaurant food is already a low one.
Raw versus cooked is the sneakier one. Cooking drives water out of meat, so a given weight of cooked chicken carries meaningfully more than the same weight of raw — about a third more. Six ounces of roasted chicken breast runs about 280 calories with 53g protein. Apply raw-weight numbers to that same six ounces sitting on your plate and you'd log roughly 205 calories and 38g protein instead. That's 15g of protein you ate and didn't count, and it isn't a one-time slip — every chicken breast after it leans the same way.
That one is worth more than any amount of care about scoop sizes, and almost nobody mentions it, because it isn't a discipline problem. It's just an easy thing to get wrong quietly.
Mark it, so the log doesn't lie about itself
The other half of estimating honestly is that the log has to admit which of its numbers are soft.
Every entry Graham writes carries a precision: exact when it came off a label or a known product, approx when it's a food he knows and a portion you eyeballed, estimate when he built the numbers out of your description. It isn't a note in the margin — it's a required field on the log, so there's no such thing as an entry that forgot to say.
Estimates render with a tilde on the figures and a small badge. Your Tuesday shows you which meals were measured and which were called, and a week made mostly of tildes tells you something a week of clean-looking numbers would hide.
The alternative is what most trackers do, which is to render a guess and a barcode scan identically, at the same decimal place, and let you believe them equally.
Nobody's going to chase you about it
If the real number turns up later, you say so and the estimate firms up. The label said 640 calories, the patty was six ounces and not eight, whatever it is — that clears the tilde and the badge along with it.
What won't happen is Graham asking. He doesn't come back to nag you about the casserole, doesn't ask you to be more accurate next time, and doesn't end the conversation by asking what else you've eaten today. A rough log now beats a precise log never, and treating vague input as welcome input is the only way the vague meals get logged at all.
What a fuzzy day is worth
There's a real question underneath all this, which is whether estimated days are worth counting when the numbers get used for something.
Graham's answer is that they count fully. A day built out of estimates still goes into the trend math, because fuzzy data is still data and the alternative is pretending Thursday didn't happen. His language just gets appropriately vaguer when he talks about it.
The days that get held back are different: the ones that look half-empty. If a day has almost nothing logged, treating it as though you ate 1,100 calories isn't honest either — that's not a light day, that's a day you stopped logging at lunch. Those get shown to you normally and kept out of the adaptation math.
Estimated is not the same as missing. One of them is a slightly blurry fact and the other is a hole you'd be filling with fiction.
What I'm not telling you
An estimate isn't a measurement, and I'd be wary of anyone who blurs that. If you're peaking for something, or working with a coach who needs real numbers, or genuinely close enough to your target that a couple of hundred calories decides the week, weigh your food. The scale is better than the guess, and it always will be.
The rest of the time you're choosing between a rough log and no log, and that's a much easier call than tracking culture makes it sound. Most of what you need out of a week is the shape of it: roughly what you ate, roughly when, roughly how much protein. A tilde in the middle of that costs you very little.
Your imperfect Tuesday is worth more than your perfect nothing. It's the only one of the two that exists.
— Ryan