How EatSure grades food

Every scan opens with one of four levels. It is coloured, it comes before any explanation, and it is never a number.

There is no 0–100 score and there will not be one. A single number invites you to compare two products that are not comparable, and it hides which of a dozen different things produced it.

The same cookie is different answers to different people. To someone with a milk allergy it is Not for you. To someone cutting added sugar it is Go easy, and the reason names the sugar. To someone who has told us nothing it is Go easy too — but for a different reason, and one that would be true for anybody: the box is most of a day's added sugar. A rating with no subject is mostly an opinion about food, which is why an empty profile can never reach Great for you.

That is why every result names who it was computed for, and why the top level stays locked until you tell us something. We cannot say a product is great for you before we know what you are looking for.

An allergen or a dietary rule you set is absolute. If a product crosses one, it is Not for you, and nothing else on the label can outweigh it — a certified seal and a perfect nutrition panel do not soften an allergen.

Hard lines split into two kinds by one test: does crossing it harm the body? Both are red, and they are not treated the same.

The allergen kind is the nine major allergens US law requires on every pack, plus gluten. Gluten used to sit with the dietary rules and does not any more: coeliac disease is a physical reaction, so it gets allergen-grade care rather than lifestyle treatment. It is worth knowing that gluten is the one in this group nobody has to declare — wheat must be named, but barley and rye carry gluten under no labelling law at all, so we lean harder on the ingredient list for it than for the other nine.

The dietary kind is vegan, vegetarian, halal and kosher. Still red when crossed, and still yours — but a choice rather than a physical limit, which is why these carry the softeners further down: a certification mark can answer one of them, and we say out loud that ingredients alone cannot prove how something was manufactured.

You can also type an allergen we do not have a checkbox for — pumpkin, mango, mammalian meat for alpha-gal. We match what you typed against the printed ingredients as well as we can, and a match is red exactly like any other. What we will not do is call a miss a pass: a typed allergen is not on the legal “Contains” line, so a name match cannot see hidden forms or shared equipment, and when we cannot confirm one is absent the result says so instead of showing a clean bill of health.

That is the one place we deliberately hold a rating back from being reassuring. Missing an allergen is the worst mistake this app can make, and an honest “we could not check” costs a line where a wrong “all clear” could cost a lot more.

Three other things land in the same place: an ingredient regulators have restricted, which is true for everyone rather than a preference; a product currently under recall; and an ingredient we hold back specifically from children, if you told us you are shopping for one. Each says which it is — a red for a child does not read the same as a red on your own hard lines, and it would be wrong to tell you regulators restricted something when nobody did.

We also tell you what a pack declares even when none of it is yours. “For the record it contains milk and wheat” costs one line and is the sentence somebody shopping for another person needs.

Saturated fat, sodium and added sugar are the three the FDA itself labels as ones to limit, and the agency publishes the line: 20% of a Daily Value in what you eat is “high”. When a product crosses it we say so, whether or not you have told us anything about yourself.

We used to keep quiet unless you had named the nutrient, on the reasoning that a number is a fact and only a stated goal makes it a judgement. We changed our minds. “All clear” on a pack carrying a fifth of a day's saturated fat reads as an endorsement, and this is not our threshold to soften — it is the one printed on the side of the box.

It moves a rating to Go easy and no further. Nothing but the hard lines above reaches Not for you — no nutrient, at any amount, ever does.

Total fat is measured and never judged. It carries a HIGH tag on the panel when it clears the same 20% line, because that is a fact worth seeing, and it moves no rating on its own — a high total fat can be perfectly good fat, and olive oil, nuts and avocado would all be wrong to flag.

Trans fat is the one exception to having a line at all. There is no amount that counts as low, so if a label lists any, we mention it. That also means we can only see what a label admits to: US rules let a pack print 0 g when a serving holds less than half a gram.

A panel prints one serving. People eat packs. A six-serving snack at a fifth of a day's added sugar per serving is more than a day's worth if you finish it, and reading only the printed number would call that fine.

So when a pack is realistically one sitting, every number we judge is the pack's, and the reason says so — including the honest calorie total. A genuinely multi-serving staple is not treated that way: a bag of flour is a container you portion out of, and flagging it because its numbers add up over thirty servings would be true and useless.

We tell the two apart from what the label prints, never from a guess about the product. A serving that counts several pieces — four cookies, about twelve chips — is a portion of something loose, and the pack is the portion. A serving measured in cups or grams is something you spoon out of a bigger container.

A serving of exactly one wrapped thing is the third case, and it is neither: a box of six bars is six sittings, not one, however countable a bar is. Scaling those together would have put every multipack in the shop on Go easy, which is the fastest way to teach somebody to ignore a rating.

Calories work the same way and never as a percentage. The 2,000-calorie figure on a label is the reference the other percentages are computed against, not your budget, so we will say “the whole pack is 1,020 calories” and never “half your day”.

Calories also move a rating for two goals only, and in opposite directions. If you are losing weight, a calorie-dense food or a high total for a pack you would finish nudges toward Go easy. If you are building muscle, calorie density counts as a fit alongside protein rather than against it. For every other goal, and for no goal, calories are a number on the panel and nothing more.

The rule above cuts the other way too. Soy sauce carries about two-fifths of a day's sodium in a tablespoon, and nobody sits down to a bowl of it. Judging that number the way we judge a ready meal would put every condiment in the shop on Go easy — which, again, teaches people to ignore ratings.

So when a pack's own panel says it is spooned out — a serving measured in teaspoons or tablespoons, weighing 15 grams or less, out of a container holding at least eight of them — we stop counting its per-serving nutrients against your goals and against the three universal limits. Everything else still applies: an allergen is still an allergen, a restricted ingredient is still restricted, and any trans fat is still reported.

The panel keeps its HIGH tag on the number either way, and the reason line names the nutrient and the amount the pack is measured in: “This is high in sodium. Try to keep it to about a tablespoon a meal.” The amount is the label's own serving size, not a figure we chose. An unexplained missing warning is worse than a warning.

Two cases drop the amount. When the pack is measured in a fraction of a spoon — a salt canister printing a quarter teaspoon — naming a whole teaspoon would suggest four times the serving the number was printed for, so the line says the nutrient is high and stops there. And when something on the label rules the product out for you outright, there is no portion advice at all: a page that has told you not to eat something does not go on to say how much of it to use. The number is still stated as a plain fact.

That sentence is the one place this app suggests how much of something to have. Everywhere else it states what a pack contains and leaves the decision alone. It is also the reason the rule is read off the serving size and never off a guess about what the product is: a rule that quiets a warning has to rest on something the pack actually printed.

We read this off the printed serving size rather than off a description of what the product is, and that is a deliberate limit. A rating that silences a warning must not rest on a guess about the food — so a pack has to say it is spooned out, not merely fail to look like a meal. A tablespoon of cooking oil qualifies; a small biscuit weighing the same does not.

A goal can move a rating down one level, to Go easy — never to Not for you. A goal is something you are working toward, not a line you must not cross, and treating the two the same would make the serious one meaningless.

A goal changes which nutrients we watch, not how high counts as high. There is one line for that and it is the FDA's: 20% of a Daily Value, in the amount you actually eat. We ran two for a while — the agency's, plus a lower one for nutrients you had named — on the reasoning that naming a nutrient means it matters more to you than to the average shopper. It does, and the honest way to act on that is to watch more nutrients for you, not to answer “is this a lot?” two different ways on one page.

A goal watches every nutrient it is about, not one. Heart health watches saturated fat and sodium; managing cholesterol watches saturated fat and dietary cholesterol.

A goal can also watch an ingredient, which so far means one: alcohol, for heart health and for liver health. Alcohol has no Daily Value and never appears on the nutrition panel, so there is no percentage to quote — the reason names the ingredient instead. This is separate from the halal rule, which is red for anyone who set it regardless of goals.

It has to be alcohol the product is actually made with, not a trace of it. A splash of cooking wine near the top of an ingredient list counts; ethanol used as a solvent inside a flavouring listed near the bottom does not. We read that from where it sits in the printed list, which US law requires to run from most to least by weight — so “second ingredient” means something and “fourteenth” means something else.

Nesting is judged by the component it sits in rather than by the ingredient alone, because a label is required to spell out its sub-recipes: a beer batter and a wine-braised sauce both list their alcohol one level down, and refusing all of those would go quiet on exactly the foods where cooking alcohol lives. And a bottle of vanilla extract does not warn, because it is spooned out — it is legally at least 35% alcohol and names it second on the list, behind only water, which is a quantity question rather than a presence one.

We will miss some. A tiramisu or a Christmas pudding lists its marsala or brandy behind six dry ingredients, and a rule tuned to catch those would flag every trace in every flavouring. This is a nudge, not a safety line — the safety line is the halal rule, and that one fires on any alcohol anywhere.

Liver health watches added sugar, saturated fat and alcohol. Kidney watches sodium and nothing else, and the app says so in a line under the goal picker rather than leaving it buried here. Managing a kidney diet properly means potassium, phosphorus and protein limits that depend on the stage of the condition — that is clinical, a label cannot support it, and calling a sodium-only check “kidney management” would be a claim we have not earned.

Goals also work upward: a nutrient your goal is actually about, at a meaningful amount, counts toward Great for you. Fibre qualifies at the “good source” line; protein has to clear the “high” one, or ten grams outright, because a token gram was never what a muscle goal meant.

Every percentage on a nutrition label is measured against a 2,000-calorie reference person. That is a real standard and we never change it — the label's own %DV stays on the panel exactly as printed.

But if your real daily need is nearer 1,700 calories, the same 20% is a bigger share of your day than of that reference person's. If you tell us your height, weight, sex and birth year, we can show that second number beside the first, on the five nutrients whose recommended amounts genuinely scale with how much you eat: calories, saturated fat, added sugar, fibre and protein. Sodium and the vitamins and minerals are flat population figures — scaling those would be inventing a number, so we do not.

It is off unless you turn it on, it is an estimate rather than a measurement, and it never changes a rating. Asking for your weight and then grading your food differently because of it would be a different product from this one.

It never shows a calorie budget, a target or a deficit — only what share of an estimated day a nutrient is. We do not offer it at all to anyone under eighteen, and the switch says plainly that if you are pregnant or have a history of disordered eating, leaving it off is a reasonable thing to do. Your body stats are never shared and are never part of anything you could pass to someone else.

We count additives and we tell you what they are for, because those are facts. For almost everybody the count is only ever a fact: there is no threshold where a number of additives becomes a verdict, because that number would not mean anything on its own. The single exception is the switch below, and it is off unless you turn it on.

A preservative does a job — it stops food spoiling. Colours, flavours, sweeteners, emulsifiers and thickeners are what heavy processing looks like. Grouping all of them under one scary number would tell you less than the count and the purposes do separately.

Cosmetic additives move a rating for one person only: someone who turned on “prefer minimally processed”. It is one switch, it is off by default, and it is the only thing that makes an additive count against a product without cited evidence behind it. When it fires, the reason says whose rule it is rather than describing the food.

It works upward too, for the same person: a pack we read that turns out to carry no cosmetic additives at all counts toward Great for you. We do need to have read it — “clean” is a claim, and a photo we could not make out is not evidence of one.

An additive with real cited evidence — a dye tied to attention in children, say — moves a rating through its reviewed knowledge-base card, like any other researched ingredient. That is the only additive signal that is a health claim, and a human approved it.

A comparison like “more additives than most frozen dinners” would be more useful than a raw count, and we do not make it, because we do not yet carry product categories to compare against. We would rather say less than imply a comparison we have not done.

A certification counts toward Great for you only when somebody actually certifies it. “USDA Organic” is checked by an accredited agent. “Natural” is checked by nobody — it is in our vocabulary precisely so we can tell you that, and it never counts as a match.

And only when it is a certification you said you look for. A seal you never asked about is printed as a fact about the pack; it is not a point in the product's favour, because we would be scoring it against a preference you never expressed.

Every ingredient rating comes from a card a human reviewed, with sources you can open and read. The cards live in version control, not in a database somebody can edit quietly: a change is a reviewed pull request, and the history is public to us.

When a scan meets an ingredient we have never reviewed, we describe it on the spot and keep the description, so the next person who scans it waits for nothing. That description never sets a rating — only a reviewed card can — and it never carries an allergen or a dietary claim, because those are the ones where being wrong hurts somebody. The ingredient joins a queue for review, ordered by how many people actually met it.

Once a week we ask whether authoritative guidance has moved on the cards we already have. That check can propose a change; it cannot make one. A claim that arrives without a citation is discarded rather than recorded.

We do not invent a photo. Product images come from the recall notice, your receipt, or the picture you took. Ingredient photography is generic on purpose.

We do not guess a verdict. We will describe an ingredient we have not reviewed yet — a dead end that says “still researching” helps nobody standing in an aisle — but that description never sets a rating, never claims an allergen, and never claims a dietary property. Where a pack contains ingredients we have not reviewed, the result says so and says we could not confirm them against your allergens.

We do not rate a photo we could not read. If the nutrition panel or the ingredient list is blurred, glared out or covered, we ask for another photo instead of producing a verdict — either one of them, because each blinds a different half of the answer. A rating with nothing behind it would read as All clear, which is the most reassuring thing we could say and the least earned. A pack that genuinely prints no ingredient list, like a bottle of olive oil, is a different case and still gets its rating.

We do not produce a score. No grade, no percentage, no traffic light standing in for a paragraph we could have written.

We do not sell any of this. Your profile decides your ratings and stays yours.