Two baskets roll down the belts. You call which one is healthier. The game learns what you believe about food, then holds it up to Sanctity’s nutrition reference.
14 baskets · about 3 minutes · no wrong answers until the end
A Sanctity AI Lab experiment.
SanctityHealthy Basket
1 of 13
Called 0/0
I'm Pixel. First time here? A 30-second tour.
Pixel explains Healthy Basket
Each item is one everyday serving.
You have the hang of it. Go express? Faster belts, no camera moves.
Lab notes on you
The game forms a hunch, tests it with a few baskets, then moves on. Sanctity’s nutrition reference comes at the end.
Belts paused
Your baskets so far are kept for a day. Pick up where you left off.
Every basket you bought, with the nutrition reference's call beside it. Tap a line to see why.
The reference's call adds up each food's place in the evidence, one everyday serving each. The reference is the Sanctity Lab's reading of Nutri-Score 2023, NOVA processing groups, WHO sugar guidance and IARC. For learning, not dietary or medical advice.
SanctityHealthy Basket
Tap to go on, tap the left edge to go back←→ move · ↓ full profile
SanctityHealthy Basket
Your food profile
match with the nutrition reference on the foods you weighed
of our sealed guesses came true
average time to decide
You said you were guided by
Your choices say
Where you part from Sanctity’s nutrition reference
Sanctity’s nutrition reference: the Sanctity Lab's reading of Nutri-Score 2023, NOVA, WHO and IARC, one everyday serving each. For learning, not dietary advice.
YouNutrition referenceSanctity players
Your blind spots
Foods you weighed where your ranking and the reference's ranking sit furthest apart.
Your full ranking
Each row is one food, listed from your healthiest down. Across the row, left is less healthy and right is healthier: the orange dot is where you put it, the faint band is how sure the model is, and the green ring is where the reference puts it. Hover or tap a row.
Your tiers
Foods the model cannot yet tell apart share a tier.
How fast your profile took shape
How close each basket's ranking was to your final one, and how much each basket taught the model, in bits. Tall bars were the close calls.
Create a free account to see more
Your full ranking of every food, your tiers, how fast your profile took shape, and Sanctity players next to you. Kept with your account.
Every food has a weight in your head: how healthy you think it is. A basket's weight is the sum of its foods. Your picks train a small statistical model of those weights, and of seven beliefs underneath them: how much fat, sugar, processing, starchy carbs, protein, fruit and veg, and a health label move your sense of healthy.
The next basket is not random. The first three explore. Then the game forms a hunch about one belief, tests it with baskets that differ in that one thing and match on the rest, and moves on once it is sure. Once, it brings back an earlier pair with the sides swapped, to see if you call it the same way. The last baskets are close calls on the foods where you and the reference seem to part.
How long you take counts too. A very fast click counts a little less, and so does a very slow one. "About the same" counts as a tie. If time runs out, the basket barely counts.
The reference markers are the Sanctity Lab's read of nutrition science for a typical version of each food, one everyday serving: Nutri-Score (2023 algorithm), the NOVA processing groups, WHO guidance on free sugars, and the IARC classification of processed meat. They are approximate, and nutrition science is still moving. The reference match counts the pairs of foods you weighed where you and the reference put the same one on top.
The method is BasketSort, Sanctity's preference engine: forced choices between bundles, a Bayesian model of what you value, and each next question picked to learn the most.
Healthy Basket is a Sanctity AI Lab experiment. It is not dietary or medical advice.
How Healthy Basket works
You are at a supermarket checkout. Two baskets of groceries roll down two belts. You say which basket is healthier, and that basket goes through the scanner. There is no right answer during the game, only yours. Each item is one everyday serving.
The game adapts as you play. It forms hunches about what you believe (is fat bad? is honey better than sugar?) and picks each basket to test them, so fourteen are enough for a first, provisional pattern. You can stop early, or ask for more baskets to sharpen it.
At the end you see where you part from Sanctity’s nutrition reference, food by food and belief by belief, and how you compare with other Sanctity players once enough have played.
Where the reference comes from. Each food's place is the Sanctity Lab's reading of public sources for one everyday serving: Nutri-Score (2023 algorithm), the NOVA food processing groups, WHO guidance on free sugars and IARC's classification of processed meat. It is general nutrition science for learning, not dietary or medical advice. If you have a health condition, talk to a doctor or dietitian.
Foods you know. The baskets hold foods eaten where you live: the region you gave puts you in one of six food-habit groups (Europe, North America and Oceania; Latin America and the Caribbean; the Middle East, North Africa and Central Asia; Sub-Saharan Africa; South Asia; East and Southeast Asia). The Nutri-Score grades on the receipt are the Lab's calculation from typical composition per 100 g (most from USDA FoodData Central, the Indian Food Composition Tables, UK CoFID, the FAO West African and Kenyan tables, Japan's food tables), not product labels. Sanctity is not affiliated with any of these organisations. The world map is drawn from Natural Earth (public domain).
Setting up the next basket
The connection dropped
Sanctity AI
AI has no values of its own. It inherits ours.
The values layer for AI: a public AI constitution, chosen by people’s answers.
The Constitution
Eight Articles, written in public.
Sanctity writes the candidate lines; public answers help decide which are proposed. Every line names the question it came from.
