How Nibavo Works

Photos, AI and evidence

When you permit cloud analysis, selected prepared photos are uploaded through our private processing service and interpreted using OpenAI. The app then organizes the result, applies your profile locally and calculates supported nutritional indicators locally. AI output is an interpretation of the supplied evidence, not manufacturer verification.

The app can show:

• Read from your photo: a label value or statement linked to the photographed evidence; it can still be misread.
• Calculated from label values: an arithmetic conversion or nutritional indicator using stated inputs and units.
• Estimated or general information: an inference from visible features or product type, not a confirmed ingredient, nutrient amount, expiry date or safety finding.
• Not available: evidence is missing, inconsistent or unsuitable for the requested conclusion.

A partial label may support useful identification or ingredient explanations without supporting a numerical nutrition score or an allergen conclusion. Another photo can improve evidence. An estimate is not a value read from your label and must not be treated as a measurement.

Your profile is a separate check

Your optional food profile is evaluated locally against the available product evidence. A possible restriction match deserves attention regardless of a product’s nutritional score. “Not found in the available information” is different from “does not contain”. Cross-contact and unreadable precautionary statements cannot be ruled out by a favorable result.

Halal, kosher, vegan and other dietary practices have rules that a label or photo may not establish. Nibavo does not certify compliance. Your own fasting preferences do not turn a product analysis into a determination of a religious obligation.

Score

Nibavo’s 0–100 nutritional indicator is our own transparent method. The current policy is internally identified as foodscanner-nutritional-balance-v1; the historical technical name does not identify another operator. It is not Nutri-Score, Yuka’s formula, a health probability or an official certification. It does not measure how safe food is for an individual.

The method uses supported label-derived sugar, saturated-fat and salt information, with the policy’s protein/fibre contributions and energy context. Different food/drink contexts and units matter. Open the score explanation to see the active method version, selected nutrition table, starting points, each positive/negative contribution and the resulting total. The calculation schedule and illustrative example below describe this method.

A complete score requires the relevant trustworthy inputs. Missing inputs are not zeros and must not be invented from the brand or another product. The app may show partial factors without an overall number. Water and unsupported/special categories are handled under the method’s exclusions; a missing score is not a negative verdict.

As-sold powder and a prepared drink are different contexts. We do not mix rows from incompatible tables or assume the amount/type of milk added. A serving-to-100 g/ml conversion needs a known serving mass or volume; grams and millilitres are not interchangeable without appropriate evidence.

Current calculation schedule

Use grams of each nutrient per 100 g of food or 100 ml of drink and energy in kcal on that same basis. Start at 60 points. Round each contribution separately to the nearest whole point, with half-points rounded away from zero, then add the signed contributions.

For each negative factor, the penalty is zero at or below the lower reference, increases linearly between the references, and reaches its maximum at or above the upper reference. These are mathematical reference points; they do not by themselves authorize a regulated nutrition claim or an official traffic-light label.

For reproducibility, let C(x) = min(1, max(0, x)); S, A, N, P, F mean sugars, saturated fat, salt, protein and fibre in grams, and E means kcal:

sugarPenalty = round(25 × C((S − lowerSugar) / (upperSugar − lowerSugar)))
saturatedFatPenalty = round(20 × C((A − lowerFat) / (upperFat − lowerFat)))
saltPenalty = round(15 × C((N − lowerSalt) / (upperSalt − lowerSalt)))
proteinPoints = round(20 × C((100 × 4 × P / E) / 20))
fibrePointsFood = round(20 × C(max(F / 6, (100 × F / E) / 3)))
fibrePointsDrink = round(20 × C((100 × F / E) / 3))
score = 60 − sugarPenalty − saturatedFatPenalty − saltPenalty
        + proteinPoints + applicableFibrePoints

If printed energy, protein and fibre are all zero, positive contributions are zero rather than dividing by zero. Inconsistent or unsupported inputs are withheld rather than forced through the formula. Salt may be derived from eligible printed sodium by multiplying by 2.5; eligible kJ may be converted to kcal by dividing by 4.184 when kcal are not available. These conversions are identified rather than presented as newly measured values.

Illustrative calculation, not a scanned product: for a supported food table per 100 g with 200 kcal, 10 g sugars, 2 g saturated fat, 0.3 g salt, 8 g protein and 4 g fibre, the contributions are −7, −3, 0, +16 and +13. The result is 60 − 7 − 3 + 0 + 16 + 13 = 79/100. This number says nothing about an individual allergy or whether the item has been stored safely.

The weights are Nibavo’s product-design choices and have not been clinically validated against health outcomes. Allergens, freshness, religious certification, taste, processing, micronutrients, typical portion size and eating frequency are not hidden inputs. Scores are not assigned to pet food, infant/medical nutrition, supplements or non-food items under this method. Color bands, where shown, are our own display intervals, not an official health classification.

Comparisons and opinions

Comparisons use eligible products you previously scanned, on the displayed common metric and basis. They are not a search of all shops or a promise to find the healthiest product. A lower sugar bar establishes only that displayed difference; it does not override allergy conflicts or make an alternative universally preferable. Likes/dislikes reflect your preference, not clinical validation.

The current scoring implementation contains no paid-ranking or sponsored-placement input. Introducing sponsorship or affiliate recommendations would require clear disclosure and a review of how recommendations remain fair.

Using the result

Use the summary to decide what deserves a closer look. Check the physical package where a restriction, medical need or uncertainty matters. A photo cannot establish contamination, spoilage or a missing expiry date. Please contact support about a misread result without sending unnecessary sensitive information.