How we score
Last updated 27 October 2026
The short version. Five signals, each worth up to 20 points, added to a total out of 100. They are read from the label, plus what we know of the brand and category. A separate Material Score reads only the physical package. Every result carries a confidence, and the limits are stated below rather than buried.
The Greenspecs Facts panel is a different thing: a factual label, not a verdict. Its numbers never move the score.
This is the same explanation as the How we score screen in the app, with the Facts panel added. Nothing here is a summary of something longer that we keep private.
The five signals
| Signal | Worth | What earns points |
|---|---|---|
| Claims and disclosure | 0–20 | Specific, verifiable claims. "80% recycled content, third-party verified" beats "eco-friendly" every time. Vague language is cheap; published data is valuable. |
| Certifications | 0–20 | Marks with real audit requirements: B Corp, USDA Organic, Fair Trade, EPA Safer Choice. A leaf logo the company designed itself does not count. |
| Packaging lifecycle | 0–20 | The full picture, not just "is it recyclable". Refillable with a deposit scores highest, glass or not. Concentrates, bulk and a light bag sit next. Single-use glass is recyclable but heavy to make and ship, so it lands mid-table. |
| Ingredient impact | 0–20 | What is in it and how it was made. For food: fewer recognisable ingredients, less processing, organic on high-risk crops. One organic ingredient plus fifteen additives does not score well. |
| Supply chain | 0–20 | Where it was made and how it got here. Short chains, published Scope 3 data and verified supplier standards score highest. Clean ingredients shipped globally with no offset cap this signal. |
The five are summed by the model that reads the label, inside a fixed rubric. The bands that turn a total into a letter are in code, shared by the app and the store badge, so a badge and a scan can never disagree: 85+ is A+, 78 an A, 72 an A-, 66 a B+, 60 a B, 55 a B-, 50 a C+, 44 a C, 38 a C-, 32 a D+, 26 a D, 20 a D-, below that an F.
The limits, stated
A camera reads a label. It cannot see a supply chain, a farm or a factory. The supply chain signal comes from what we know of the brand and category, and the result says so.
When a label says nothing on a signal, that signal scores low, not neutral. That is a reading of how much the brand discloses on the pack, not of the product's impact. The result names what the label does not say.
A few ingredients cap the total whatever the rest says: phosphates, NTA, nonylphenol ethoxylates, and chemical sunscreen filters such as oxybenzone. Well-known products are held against reference scores so the same product reads the same each time.
The Material Score is separate. It reads only the physical package in the photo. No claim on the label moves it, and a missing ingredient list is a data gap there, not a penalty.
Every result carries a confidence: high, medium or low. It drops one step once a scan is older than 180 days, automatically, at the moment it is read. A 74 at high confidence and a 74 at low confidence are not the same 74.
A grade cannot be bought, edited or hidden — by a brand, a store, or us. A brand can send evidence, such as a certificate; we check it before it counts. A claim we cannot find in an issuer's own register never lowers a grade; it only fails to raise one.
The Greenspecs Facts panel
A nutrition-facts-style sustainability label rendered for every scan. It states facts and comparisons. It does not rank, recommend or endorse, and it never replaces the score card.
Greenspecs defines this schema. External data sources — EU Digital Product Passports, manufacturer disclosures, certification registers — are inputs that may fill or upgrade a field. They never define field names, structure or methodology.
Source tiers
Every metric on the panel is a triple: a value, a unit, and a source tier.
| Tier | Meaning | Who can assign it |
|---|---|---|
| verified | Third-party-verified or regulatory-grade data | The ingestion seam only — external feeds mapped into our fields |
| claimed | Manufacturer-disclosed and unaudited: printed on the pack or published in the brand's own reporting | Scan analysis |
| estimated | A Greenspecs category average, or modelled from established lifecycle benchmarks | Scan analysis |
- The model may estimate a value, and must label it estimated. It can never emit verified: any such output is downgraded to estimated in code before it is stored.
- Estimated values are marked on the panel with a ~ and a footnote.
- A field with no defensible value is empty and reads not disclosed. Null is often the correct answer.
- Tiers only ever upgrade: estimated to claimed to verified, never the other way. When a verified value replaces an estimate, the change is logged with the old and new value, the source, and the score before and after, so any movement stays explainable.
Field definitions
All CO2e and water figures are per purchasable unit — the thing you actually buy, such as a 12-roll pack — across the full lifecycle.
| Field | Definition |
|---|---|
| Carbon total | Full-lifecycle greenhouse-gas footprint, kg CO2e per unit |
| Scope 1 | The share from the maker's direct operations |
| Scope 2 | The share from purchased energy |
| Scope 3 | The share from the supply chain. Scopes 1, 2 and 3 sum to the total |
| Fossil free | Percentage of direct operations running without fossil fuels |
| Renewable | Percentage of purchased energy from renewables |
| Scope 3 disclosed | Percentage of supply-chain emissions the brand measures and discloses, not estimates. The panel's footnote is its inverse |
| Water use | Litres across the lifecycle, per unit |
Comparisons
- Category averages, deltas and percentiles are precomputed from the scan database on a weekly schedule. They are never computed per scan, and never by the model.
- Deltas are signed percentages against the category average. Negative means less impact. A delta needs at least five samples in that category for that metric.
- "Cleaner than N% of category" is a lookup against nine precomputed cut points in that category's score distribution.
- The real scan count is always printed. Below 30 scans the panel says so in as many words: comparisons will sharpen as scans grow. A small sample is disclosed, never hidden, and never suppresses the panel.
Certifications
The panel lists the certifications expected for the product's category. A held certification gets a check; an expected certification that is absent gets a dash. Absence is information. Claims of B Corp, USDA Organic and Leaping Bunny are checked against the issuer's own register, and a B Corp claim is checked by a person, because B Lab publishes no open data.
What the panel will not do
- No ranking language, no "better choice" badges, no endorsements — in the panel or in its shared image.
- The data zone is deadpan. Personality is confined to the footnote and the footer.
- Facts-panel numbers never influence the score or any rubric signal. Capture and scoring are separate paths.
The AI, and what it costs
A model reads every label, and that uses energy and water. We publish what it does here, what a scan costs us, and our estimate of what a scan costs the planet, with the arithmetic shown: greenspecs.app/ai.
YKO's methodology, and how it maps
YKO.Earth, our sister site, reads the record behind the package with a different six-area rubric. The two numbers answer different questions and are never converted into each other. The map between them is public: yko.earth/methodology.
If a score is wrong
Rate any result in the app, or email hello@greenspecs.app with the product name. A person reads every report, and a product can be queued for a re-score, which replaces the stored result. Facts-panel upgrades are logged as described above.