Product intelligence for the aisle

Scan anything.
Know what you're buying.

Two products, no idea which one is better? Scan either one. Calebonne turns a barcode or a photo of the label into a clear, evidence-based answer in seconds — a score computed by published rules, not by a chatbot.

iOS 1.0 is with App Review

  • Scoring food_v1.9.12 live on 4.2 million products
  • 118 additives profiled: what each can lead to, the evidence, the law
  • Every nutrient as a row with its value and where it sits for its category
  • Lab-grown? answered on every egg, meat and dairy pack

No account needed. Basic safety information is always free.

Three separate outputs

One score for everyone. One fit for you. One honest measure of how sure we are.

They are never blended. Your goals never move the score, and thin data never gets rounded up to a confident number.

Calebonne Score

Universal, category-relative, deterministic. A 0–100 rating of a packaged food or personal-care product against realistic alternatives in its own category, computed by a fixed set of published rules from the label.

Personal Fit

How the product matches the goals, diets, allergens and avoid-list you set. Shown as its own card, in its own tint. It never changes the base score.

Confidence

How sure we are about the product's identity and label data. Missing or inconsistent data lowers confidence, and when data is too thin we withhold a definitive score rather than guess.

Try a score

Tap a product. Watch the rules decide.

Six real products. Tap one to see its score and the reasons behind it.

Your score

0 / 100

Pick a product

Six real labels, one set of rules.

— Confidence Category

Tap any card to run the score. The ring sweeps, the band pops, and the top reasons show which rules fired.

    Scores come from the live app. Labels change, so always check the packaging.

    How it works

    Rules decide. AI explains.

    The score is computed by versioned, deterministic rules. Language models only extract label data, phrase the reasons and answer grounded questions. No model ever writes a number.

    1. 01

      Scan

      Point the camera at a barcode, or photograph the label when a product isn't on file. No account required.

    2. 02

      Rules decide

      Five published components — nutrient profile, formulation, processing, category performance, evidence-backed flags — combine with fixed weights. Same label, same score, every time.

    3. 03

      Evidence explains

      Every reason carries a rule ID and an evidence level. Tap an ingredient for what it does, what human evidence says, and how much you'd have to eat for it to matter.

    4. 04

      Better options

      Alternatives from the same category with measurable reasons: "12 g more protein at similar calories", "half the sodium". Never just "healthier".

    Evidence, not fear

    Every additive, flagged. What it can lead to, and how sure anyone is.

    Each additive on a label resolves to its own profile: a four-level concern tier assigned from human studies, IARC groups and what the EU, the FDA and the US states have actually done, with the concerns in plain words, the evidence level behind them and the sources. Systematic reviews and randomized trials carry the most weight; animal and in-vitro studies are context, never treated as consumer risk on their own.

    • Red means banned or revoked by a major food authority, or a possible carcinogen by the IARC. Peach means human studies raise a concern. Yellow is worth knowing. Blue is no concern on record.
    • We don't call an ingredient toxic because it appears on a hazard list. Hazard is not risk at the dose in your food.
    • We say when evidence is uncertain, and limited evidence can only move a score a little. A banned additive caps the score at Weak.
    Browse the additive index

    Polysorbate 80

    Concern Synthetic emulsifier E433 Limited evidence

    What it can lead to

    • Thins the gut mucus layer and drives low-grade inflammation and metabolic syndrome in mice (Nature 2015)
    • NutriNet-Santé: higher intake associated with cardiovascular disease (2023) and type 2 diabetes (2024)

    What the evidence says

    The landmark 2015 mouse study showed it erodes the mucus barrier, and large French cohorts since associate polysorbate intake with heart disease and diabetes. Cohorts cannot prove cause, so the evidence is limited, but the signal is consistent.

    Regulation

    Approved in both (EFSA group ADI 25 mg/kg, 2015).

    EU · approvedFDA · approved

    Affects the score via

    FOOD_ADDITIVE_RISK_01 · −6

    The additive index

    Every additive we know, explained.

    The same profiles the app shows when you tap a chip: tier, what it can lead to, the human evidence, the IARC group, what the EU, the FDA and the states did, and the sources. Reviewed October 2026.

      Dose matters: these notes describe the evidence at ordinary dietary exposure, not a verdict on one serving. General product information, not medical advice.

      Three ways to read a label

      A number without a why is a guess. A pass/fail without numbers is a mood.

      Here is how the three apps people already use describe their own methods, next to ours. Nothing here is a claim about anyone's accuracy; it is what each one chooses to measure.

      What it doesYukaBobby ApprovedScoutCalebonne
      The verdict0–100, four bandsApproved or not1–100, category-relative, three bands0–100, five bands, plus a separate "How natural" meter
      NutritionNutri-Score, 60% of the score; the same formula for every foodNot scoredCategory-relative; NOVA 4 processing caps the score at 70, NOVA 3 at 85Scored against the product's own category: a protein bar against protein bars, olive oil against oils
      AdditivesFour risk levels; one high-risk additive caps the score at 49Over 100 flagged ingredients; the list is not publishedFixed rules (seed oils always harmful, stevia neutral); no public per-additive tiers118 profiled additives with a four-level tier, what each can lead to, the human evidence level, IARC group, and what the EU, FDA and states did. Banned or revoked ones cap at Weak
      Seed oils, "natural flavors"Not a criterionAutomatic failSeed oils always penalisedPosition-graded deductions: a leading seed oil costs more than a trace; flavoring in the top five costs more than at the end
      Evidence shownA sources pageA short "why" per ingredientA methodology page; sources not linked per flagAn evidence grade on every flag, with linked sources, and "limited" means limited
      AlternativesSame category, anywhereApproved products, store listsSame aisle with prices; every swap is an Amazon affiliate listingThe same kind of product, a real step up, ranked by what your store carries (live inventory where available)
      Whole foodsNuts, cheese and butter score on Nutri-ScorePass if the ingredients passCategory-relativeScored as what they are; the "How it's made" card reads the label's claims, including lab-grown
      Lab results——Disclosed on the product, not scoredPublic third-party findings (heavy metals and more) shown on the exact product, with source and date, never folded into the score
      PricePay what you want for premiumFree, brand-affiliatedHard paywall after onboarding; about $20 per year after a 3-day trialScores, reasons, ingredients and allergen flags free and unlimited; Pro adds fit, swaps and recall watch

      Methods as published by each app on its own help pages and store listings, read in October 2026. Yuka, Bobby Approved and Scout are trademarks of their owners; Calebonne is not affiliated with either.

      Score version history

      Every rule change, dated.

      The score version is printed on every product page. When a rule or an additive tier moves, it is listed here with the reason, the same day it ships.

      1. food_v1.9.12 · 20 Oct 2026 Drink mixes and powders are scored instead of held back: a protein powder on the protein-powder profile (per 100 g of powder), any other mix as prepared — one serving in a bottle or a cup of water, on the drink profile it is sold as — with the assumed volume stated in the notes. Better options never score below the product they replace, whatever they gain on personal fit. A label the engine rejects is now re-fetched from the maker's page instead of being reported as not this product.
      2. food_v1.9.11 · 20 Oct 2026 About 300 ingredient names added to the dictionary from a tally of what went unrecognised on the 40,000 most-opened products: yogurt culture strains named one by one, French and Spanish words on bilingual labels, vitamin forms, fruit purees and juice concentrates, enzymes and common additives. No rule changed; products that were held back as "unrecognised" now score.
      3. food_v1.9.10 · cosmetic_v1.5 · 20 Oct 2026 Allulose is scored as the rare sugar it is, not as a sugar alcohol, so an allulose-sweetened drink no longer takes the polyol cap (an Olipop Cream Soda moved 50 → 66, in line with the Vintage Cola). A perfume caps inside Good at 79. Canned beans are never offered dry beans as a swap. Search lists products with a brand before unbranded imports.
      4. food_v1.9.9 · cosmetic_v1.3–1.4 · 20 Oct 2026 Probe-driven: 300 common ingredients appended to a clean granola, 200 INCI names appended to a clean cream, 300 common product names through the classifier. Celery powder or a nitrite salt with no meat on the label is a note, not "nitrite-cured meat". "Tea tree" no longer reads as TEA; formaldehyde, resorcinol, phenol, IPBC and bare benzophenone are flagged. A perfume caps at Good: the undisclosed blend is the whole product. Canned soups, frozen pizzas and dishes, gum, jams, pickles, tortillas, plant proteins and brand-only names (Twinkies, Yoplait, Dave's Killer Bread) file where they belong; sanitizers and OTC drugs leave the skincare aisles.
      5. food_v1.9.8 · cosmetic_v1.2 · 20 Oct 2026 The dictionary audit continued across every entry: honey variants had resolved to corn syrup, plain starches (rice, arrowroot, pea) to the emulsifier entry while modified starches sat with refined flour, "beet juice" and "vegetable juice" were colour additives even inside a juice, shea took the seed-oil grade, malt extract hid as a non-sugar, and sprouted wheat, wild rice and cracked wheat were not whole grains. Each now resolves to what it is. On skincare pages the named fragrance allergens are one row listing them all instead of one identical row per allergen.
      6. food_v1.9.6–1.9.7 · 20 Oct 2026 Cooked, canned and pouched grains and legumes (40–200 kcal per 100 g) are judged on dry-equivalent protein and fibre against the aisle's dry peers, so a low-sodium canned black bean is no longer "low protein" next to the same beans dry. The phosphate entry of the ingredient dictionary had carried the generic words "acid", "acids", "humectant", "silica", "calcium chloride" and "acidity regulator": every "X acid" (stearic, oleic, acetic, hyaluronic), every firming or anti-caking salt and every EU functional name had been counting as a phosphate, a red-tier ultra-processing marker, on labels with no phosphate at all. Each now has its own neutral entry; true phosphates are unchanged. Search lists a photo-only twin with no label after the scored product, never above it.
      7. food_v1.9.1–1.9.5 · 18–19 Oct 2026 Brand-by-brand audit of what the Erewhon aisle actually buys: every change fixes a correct label the engine was getting wrong, none loosens a rule for junk. A splash of sugar diluted by a 355 ml can, or explained by a listed sweetener and matching calories, is no longer withheld as a contradiction (Poppi, Liquid Death); a stored NOVA 4 is not believed for a whole-food list whose only marker is a natural colour (Simple Mills); per-can calories and caffeine stored as per-100 ml are rescaled (Olipop); label footnotes and "1% or less of:" openers are stripped before anything reads them (every Chobani Zero Sugar had been withheld); protein-by-design snacks and fibre-first candy pass the aisle ceilings they break on purpose (Whisps, Quest, Catalina Crunch, Lily's, SmartSweets); cacao-first chocolate may be two items long and "chocolate liquor" first counts as cacao (Taza, Tony's); cheese-flavoured snacks are no longer filed as cheese; grouped labels ("Crust (…), Sauce (…)") are not short lists; broths may sit under the soup calorie floor. Full catalog rescored 19 Oct.
      8. App 1.0.1 · 16 Oct 2026 The product page reads in two lists under the score: Love this and Good to know, one row per fact with its value per serving (or per 100 g), a pastel meter showing where it sits between the category's low and high marks, and the engine's own sentence a tap away. Every core nutrient has a row whether or not it moved the score; the Additives row counts every flagged ingredient, never "none" over a list; "Lab-grown?" is a row on every animal product. Ingredient and additive sheets open at one height and scroll inside. No rule changed: same food_v1.7 scores.
      9. food_v1.9 · 16 Oct 2026 Accuracy pass after a second set of four aisle audits that checked what the app shows, not only how it ranks. Soups served in ml are no longer judged on drink sodium anchors; absolute bands now apply whatever the aisle (sodium 900 mg+ is high, "low sodium" only under 140 mg; added sugar 15 g+ high; saturated fat 15 g+ high). A panel the product cannot have is not scored: 0 g saturated fat on real fat is a blank, calories that the protein, carbohydrate and fat on record cannot produce are another product's panel, fibre has a ceiling per aisle and sweets a real calorie floor. A strength needs a source — "high fibre" only with a fibre-bearing ingredient listed, "high protein" only with a protein food in the first three — and a 5 g serving earns none. A list that cannot explain its own panel (sugar with no sweet ingredient, salt with no salty one, a breaded product with a one-item list, a list that reads "in a savory sauce…") is withheld. Ultra-processed formulations are never Good (cap 69); TBHQ and synthetic dyes cap at 49; three or more highest-concern additives cap at 39; any hydrogenated oil caps at 59; refined sweeteners and mechanically separated meat are concerns on their own; deli meat tops out at 85, a sweetened 15-ingredient loaf at 74, a salt-based seasoning by its sodium. Olive oil without a virgin claim is refined. Allulose, OCR'd erythritol, bare E-numbers, plural phosphates, lactates and beef-protein isolates are read correctly; the grain amaranth is no longer the dye E123. About sixty misfiled products were moved out of aisles they topped.
      10. food_v1.8 · 16 Oct 2026 Strictness pass after four aisle-by-aisle audits of the live rankings (37 aisles, every top-40, judged against Yuka and Bobby Approved). A strength is never paid on a figure the ingredient list contradicts: no "low sodium" at 0 mg with salt listed, no "low saturated fat" at 0 g under cocoa butter or cream, no fibre credit on juice, milk, cheese, oils or drinks. "Low added sugar" is restraint, not substitution: never awarded when the sweetness comes from sugar alcohols, allulose, stevia or sucralose, or when sugar sits in the first three ingredients. A formula of four or more engineered ingredients (isolates, processed fibres, sweeteners, gums) can never score Good (cap 69; 59 with a high-intensity sweetener). "Minimally processed" needs a short, additive-free list. BHA/BHT cap the score at 49 (Yuka's red), phosphate additives and caramel colour are concerns. Cooking oils are graded by their grade (virgin and cold-pressed up, refined, light and blends down). Mechanically separated meat is no longer a whole food. Ice cream, candy, cookies, sodas and deli meat can never read "minimally processed". Seventy kinds of misfiled products (tempeh in cookies, acai bowls in ice cream, fries in frozen meals, fish in deli meat) are kept out of aisle rankings. A solid that is half sugar by weight, even fruit sugar, can never be Excellent (cap 84; 79 at 60 g), and a sweet-aisle label that names sugar while its panel shows none is withheld as the wrong panel. A label whose ingredient list contradicts the product's own name (a honey ice cream carrying a chocolate-almond list) is withheld as another product's label until the pack is photographed.
      11. food_v1.7 · 15 Oct 2026 Integrity pass after a catalog-wide audit: a score is withheld when the label omits a nutrient the category weighs (saturated fat, sugar or sodium) or when a short ingredient list cannot explain the sugar on the panel; a single emulsifier in a short list no longer reads as ultra-processed; cured meat snacks are judged as deli meat whatever aisle a store filed them in; rankings inside a capped band follow the underlying composition.
      12. food_v1.6 · 10 Oct 2026 Additive intelligence: 118 additive profiles with a four-level concern tier, what each can lead to, the evidence level, IARC group, and EU / FDA / state regulation. Additives banned or revoked by a major authority cap the score at Weak; the rest take small tiered deductions. New "How natural" meter beside every score. Swaps must be the same kind of product and a real step up, never worse on the driver that sank the original, and on US shelves. Public lab findings shown on the exact product, never scored.
      13. food_v1.5 · 9 Oct 2026 Calibration to the middle ground between Yuka and Bobby Approved on a 98-product basket: leading refined seed oils, "natural flavors" by position, glutamate enhancers, phosphates and synthetic preservatives take bounded deductions; production claims printed on the label earn a small bonus; zero-sugar drinks built on high-intensity sweeteners cap at Weak; carrageenan and synthetic dyes cap at Weak. Lab-grown status answered on meat, eggs and dairy.
      14. food_v1.4 · 8 Oct 2026 Formula integrity: the position-weighted share of whole and minimally processed ingredients replaces a first-five count; isolates, processed fibre and sweetener stacks are scored as engineering; a whole-fruit lead with no added sugar is read as fruit, not as sugar.
      15. food_v1.3 · Sep 2026 Data gates: a score is withheld, never guessed, when the label is missing, contradictory or unreadable. Drink caps by sweetener. Category profiles for 47 aisles.

      Full rule list with parameters, evidence references and rationale: the methodology endpoint. Brand corrections: support, acknowledged within two business days.

      Category-aware

      Olive oil is judged against oils. A protein bar against protein bars.

      No single formula pretends a granola bar and a soda are the same thing. Each category has its own anchors, and nutrients intrinsic to a category — fat in oils, salt in cheese — are judged where it makes sense.

      80Good

      Terra Nova (sample)

      Extra Virgin Olive Oil

      vs. cooking oils

      Mostly unsaturated fat: only 14 g of 100 g fat is saturated. Minimally processed.

      74Fair

      Summit Fuel (sample)

      Protein Bar

      vs. protein bars

      High protein: 19.8 g per serving. Sweetened with an artificial sweetener instead of sugar.

      Terra Nova and Summit Fuel are fictional sample products from the Calebonne test set, not real brands. Scores use methodology version food_v1.9.12.

      Calebonne Pro

      Depth, never gatekeeping.

      Pro answers the one question that matters at the shelf: is this safe for me. Your goals, diets and avoid list on every scan, unlimited search, every better swap, offline scanning. The score itself never changes.

      Basic safety information is always free.

      Scores, key reasons, ingredients and allergen flags stay visible on every plan. Pro is about depth, never about hiding what matters.

      Free

      Everything that matters

      • Barcode scanningUnlimited
      • Score, reasons, ingredientsAlways
      • Search by name3 a day
      • Ask Calebonne3 a day
      • Better alternativesBest 1
      • History7 days

      Pro

      Everything in Free, plus

      • Safe for meYour goals, diets and avoid list, every scan
      • Search & AskUnlimited
      • Better alternativesAll of them, ranked for you
      • Offline scanningIncluded
      • HistoryForever + collections
      • Recall alertsRecall check whenever you open Home (Pro)
      Subscribe · Apple Pay or card

      or in the app through the App Store · $9.99 / month or $74.99 / year

      Billed by Calebonne, LLC through Stripe. Renews automatically until cancelled. Cancel any time: App Store subscriptions in your Apple ID settings, web subscriptions from your receipt email or by emailing support@calebonne.com. No free trial. Your Calebonne account unlocks Pro in the iPhone app.

      No pay-to-rank. Ever.

      Manufacturers cannot buy a score, a badge or a recommendation. Calebonne is funded by subscriptions.

      Rules decide, AI explains.

      Every rule has an ID, a rationale, an evidence level and a bounded effect. Changes ship as a new score version with a published diff.

      Honest about uncertainty.

      If the data is missing or contradictory, Calebonne says so and asks for a label photo instead of guessing. Read the methodology

      Questions

      Frequently asked

      Two products. One answer.

      Get launch news, early access and the occasional swap idea. No spam, no selling your address, unsubscribe in one tap.

      App Store launch in —

      App Store launch in —

      Get the app