How Dossier scores things

Every number in this app should be checkable. This page is where the working is

written down; the product pages link back to it.

The short version

A food score is 60% nutritional quality, 30% additives, 10% organic

certification. A cosmetic score is 60% ingredient hazard, 25% fragrance

allergens, 15% irritation risk. Your profile then applies **named, bounded

adjustments** on top, and every one of them is itemised on the product page.

If we don't have enough data, you get no number rather than a made-up one.

Nutritional quality (60%)

This is our own implementation of the Nutri-Score 2023 algorithm published by

Santé publique France. Nutri-Score is a registered mark and this is not an

official implementation of it — but the algorithm is public, and we follow it.

Points are counted against thresholds for energy, sugars, saturated fat and salt

(negative), and fibre, protein and fruit/vegetable/legume content (positive).

Different tables apply to beverages and to fats, oils, nuts and seeds; cheese and

red meat have their own protein rules; water is graded A outright. Non-nutritive

sweeteners in drinks carry a penalty, which is new in the 2023 revision.

We validate our implementation against Open Food Facts' own computation across

hundreds of real products. Where we disagree with them, that has to be written

down with a reason or the build fails.

The resulting point score is mapped to 0–100 using anchors derived from the same

grade cutoffs, so the number and the letter can never contradict each other.

If energy, sugars, saturated fat or salt is missing, there is no score. The

other pillars can't carry a food score on their own.

Additives (30%)

**The pillar is set by the highest risk level present, not by how many additives

there are.** A product isn't twice as risky for containing two lecithins, and a

long list of harmless thickeners shouldn't add up to the same penalty as one

substance whose approval was withdrawn. This matches how Yuka describes its own

additive scoring.

Risk comes from regulatory standing — not from an additive's reputation.

|---|---|---|

A hazardous additive also caps the whole product score at 25. Without that,

strong nutrition on the other 60% could average away a substance the EU has taken

off the market, which is exactly the kind of laundering a scoring app shouldn't

do. It also raises a flag above the score, with its source.

Additives already penalised by Nutri-Score — sweeteners in drinks — are not

charged twice.

**Anything not in our table is reported as "not assessed". That means we have no

data on it, not that it's safe.**

Organic certification (10%)

Kept to match Yuka's weighting, and labelled for what it is.

Organic certification under Regulation (EU) 2018/848 is a farming standard.

It tells you how the ingredients were grown. It is not a measure of nutritional

quality, and an organic biscuit is still a biscuit. We include it because it's

what people expect from this kind of app, not because the evidence says it

changes health outcomes. Treat it as a preference weighting.

"Is that a lot?"

A number on a label is close to meaningless on its own. Very few people can say

whether 56 g of sugar per 100 g is unusual for a chocolate spread.

So every nutrient is also compared against the other products Open Food Facts

puts in the same category — the actual things next to it on the shelf. You get

the percentile and the category's typical value: *"higher than 94% of hazelnut

spreads. Typical is 9.5 g."*

Two rules keep this honest:

• Below 12 comparable products we say nothing. A "higher than average" claim

drawn from six products is noise dressed up as insight.

• Unremarkable nutrients are dropped, rather than padding the page with rows

saying "about average".

For fibre and protein the wording flips, because being high there is good news.

This tells you how a product compares with its shelf. It doesn't tell you whether

the shelf is a good idea — a chocolate spread with the least sugar of any

chocolate spread is still a chocolate spread, and the score says so.

Cosmetics

### Is there anything carcinogenic in it?

The question people actually bring to a cosmetics scanner, answered directly

rather than folded into a number.

Ingredients with an IARC carcinogenicity classification carry it explicitly, at

the top of the page, naming the group and what it means:

|---|---|---|

An ingredient prohibited under Annex II of the EU Cosmetics Regulation caps the

score the same way. Capping only the hazard *pillar* wasn't enough: the allergen

and irritation pillars would pull a banned-substance product back up to a

comfortable-looking 43.

An uncertain (fuzzy) match can never make a claim this serious — it is reported

as "possibly contains", never as fact, and never triggers a cap.

### The table

Our INCI table covers the ingredients EU regulation actually restricts,

prohibits, or requires to be declared, plus the common inert ones so that

coverage means something. A typical cosmetic ingredient list still contains

things we have nothing to say about.

So every cosmetic page shows coverage: how much of that product we actually

assessed, weighted by position (INCI lists run in descending concentration, so

the first few ingredients matter more than the last few). Below 50% coverage we

refuse to give a number, though we still report what we found.

Matching runs exact → synonym → CAS → a tightly bounded fuzzy match. A fuzzy

match can never produce a definitive claim, counts for half weight, and never

bridges a digit difference — PEG-8 is not PEG-100, CI 77891 is not CI 77491.

Position only counts where position means something. INCI lists run in

descending concentration, so it's tempting to discount whatever sits near the

end. But preservatives, fragrance allergens and CMR substances are *always* near

the end — they're used at fractions of a percent, which is exactly why they're

regulated by concentration limit rather than by position. So a surfactant's

irritancy is discounted for being late in the list, and a restricted preservative

is not.

Fragrance allergens are a labelling fact, not a defect. They're named on the

label precisely so people who react to them can avoid them, so they only weigh

heavily if you've told us you're fragrance-sensitive.

Your profile

Two mechanically different things happen, and the difference matters.

Hard flags never touch the number. If a product contains something you're

allergic to, we say AVOID and show you why — and we leave the score exactly where

it was. Zeroing it would destroy the audit trail and tell you something false: the

product's nutritional quality hasn't changed, it's simply not relevant to you.

Condition adjustments are bounded and named. Telling us you have high blood

pressure weights salt more heavily. The adjustment is capped, it's derived from

the salt points the product already scored, and it appears on the page as its own

line with its reasoning and source. The invariant is enforced by test:

final score = base score + the sum of the listed adjustments

Every point of difference is attributable to exactly one named rule. If it

weren't, this page would be a lie.

We never show a personalised Nutri-Score letter. The letter is a registered

public-health mark with a fixed meaning; adjusting it would misrepresent it. We

adjust our own 0–100 figure and show the official grade unchanged alongside.

When we won't give you a number

• A required nutriment is missing
• Nutri-Score doesn't apply to the category (supplements, alcohol above 1.2% ABV,

infant food)

• Overall data confidence is too low
• For cosmetics, ingredient coverage is under 50%

In all of these you get a page explaining what we know, what's missing, and a

link to fill the gap in — Open Food Facts is crowd-sourced, so gaps are fixable.

Between "confident" and "refusing" there's a middle band where the score is shown

as approximate, with a hatched border and wording that says so.

Where the data comes from

Product data comes from Open Food Facts and

Open Beauty Facts, used under the

ODbL. It's crowd-sourced, which

means it's free, enormous, and uneven — which is why the completeness figure is

on every page.

Our own tables cite EFSA opinions, IARC classifications, EU regulations, WHO

guidance and clinical guidelines. Every claim in them carries a source, and the

build fails if a citation doesn't resolve or if a table hasn't been reviewed in

18 months.

What this isn't

Dossier has no medical review behind it. It reads labels faster than you can and

shows its working; it doesn't know your bloodwork, your medication, or the rest

of your diet. If you're managing a health condition, your clinician or dietitian

knows things this app cannot.