Hey Conscious – Scoring Methodology and Data Sources Notice

Draft for legal review. Text in [square brackets] is a placeholder or an item to confirm. Version 0.1, last updated [date]. Criteria, weightings and thresholds below are as at [date] and must be kept in step with the product.

1. What this notice is

This notice explains, in plain language, how Hey Conscious produces its scores and recommendations and where its data comes from. It is part of our Terms and Conditions. If this notice and the Terms conflict about how scores are calculated, this notice prevails. On everything else the Terms prevail.

It describes a method, not a promise of an outcome. Our scores are opinions based on a published method applied to imperfect data. Section 9 sets out the limits you should keep in mind.

Our data architecture. Hey Conscious is built on a Database that we designed. Every product, however its information reaches us, is mapped to the same structure: its identity (barcode, name, brand, category and country), its ingredients, nutrition and packaging, and a score and evidence note for each criterion on two sides, health (“For you”) and planet (“For the planet”). The criteria, their weightings, the thresholds, the warnings (Flags), the rules that check each record and the formula that produces the scores are all defined by us and are the same for every product. This common map is what lets us compare products with each other and find better alternatives. The Database is built to hold a very large number of products. We start with the products we expect to be scanned most often in the UK, and it grows with every new source, scan and correction. Data from outside sources and AI-assisted research are inputs to this structure. They do not define it.

2. How a score is made

When you scan a product that we have not assessed before, this happens:

• Scan. The app reads the barcode, checks that it looks valid, and asks our server for the product’s assessment.
• Check for an existing assessment. If we have assessed the product before, we show the stored assessment straight away.
• Look up open databases. Otherwise we look the barcode up in Open Food Facts and Open Products Facts for ingredients, nutrition, packaging and published ratings.
• Take published ratings. For food, three criteria are taken from published ratings where they exist and converted to our 1 to 10 scale: Nutrition from Nutri-Score, Processing from NOVA, and Production from Green-Score.
• Filling the gaps with AI research. Where information is still missing, an AI model reads the records we hold and searches the web to fill the gaps. It helps identify the product, decides whether it is Food or Household, assesses it against each criterion in section 5 (leaving a criterion blank where it cannot judge), records whether each assessment rests on evidence or was inferred, raises any warnings (Flags), and lists certifications, additives and ingredients. It also writes the plain-English summaries.
• Check. Fixed rules check that the answer uses exactly the criteria in section 5. An answer that does not fit is discarded.
• Calculate. A fixed formula, not the AI, turns the criterion scores into the two headline scores (section 7).
• Store and reuse. We store the assessment against the barcode and reuse it for later scans, until the Methodology changes.
• Show. The app turns each score into a word and a colour (section 7).

3. Where our data comes from

• Open Food Facts and Open Products Facts. Free, community-built databases of product information, contributed by volunteers and by brands. They are licensed under the Open Database Licence and the Creative Commons Attribution-ShareAlike licence. We credit “Open Food Facts contributors” and link to openfoodfacts.org.
• Published ratings. Nutri-Score, NOVA classification and Green-Score, which are produced by others. We use them as published and they carry their own limits.
• Public scientific and regulatory sources. For example EFSA, ANSES, the UK Food Standards Agency, ECHA (including the REACH list of substances of very high concern and CLP hazard classes), IARC cancer classifications, CITES, and life-cycle data such as Agribalyse.
• Web research. Where data is missing, incomplete or new, we use AI tools to search the web, including retailer, brand and certification-body websites, to identify a product and fill gaps.
• Licensed third-party datasets. Data we license from third parties, which we will list here as it is added. Each is subject to its owner’s terms, and we may not be able to publish the underlying data.
• Products and corrections from users and brands. Reviewed before use [confirm].

We do not test products ourselves. We do not guarantee that any source is accurate, complete or current. Products are reformulated, the same barcode can cover different recipes in different countries, and open and web data can be wrong. Always check the label on the product in your hand, especially for allergies.

4. Who decides what: our rules and the AI

The AI never gives the headline scores. They are always calculated by the formula in section 7. Because AI contributes to the individual criterion assessments, two researchers, human or AI, could reasonably score the same product differently. The AI is instructed to use the published ratings unchanged, but our software does not enforce this [confirm whether this is still the case].

5. What gets scored

Every product is either Food (anything eaten or drunk, including alcohol and baby food) or Household (cleaning and personal care products; anything that is neither is scored as Household). Each has criteria in two halves, For you and For the planet. Weights show how much a criterion counts within its half.

Food

Household

Household “For you” weights add up to 75 rather than 100. Only the proportions matter, which work out at roughly 47, 33 and 20.

Blank criteria are left out and the remaining weights share the whole. If nothing in a half can be scored, that half has no score.

6. Warnings (Flags)

The AI raises a Flag when an ingredient or practice meets a recognised concern. Only Flags backed by a source it found can change a score. A Flag reports that a recognised body, standard or source has identified a concern. It is not our own finding that a product is harmful, unlawful or unethical, or that any company has done something wrong.

A Flag shows how a source classifies an ingredient. It does not tell you whether the amount in a product is harmful. Classifications change, and sources can disagree. Flags can be wrong, and a Flag that is wrong can be corrected on request (section 10).

7. Calculating the scores

The same steps run for each half, in this order:

• Weighted average of the criterion scores, using the weights in section 5.
• Certification bonus. Add 0.3 for a company-level mark (such as B Corp) or 0.5 for a product-level mark (such as Organic), up to 0.5 in total. This applies to both halves.
• Long-term risk deduction. Take off 0.5 or 1 for a Fair or Poor Flag that is a long-term health risk, never more than 2 in total. This applies to For you only.
• Limits and rounding. Keep the result within 1 to 10, then round to a whole number.
• Caps. If there is a Poor Flag the half cannot exceed 4. If there is a Bad Flag it cannot exceed 2.

Words and colours. The app turns each score into one of five bands. Colours follow their own cut-offs: 8 and above is green, 4 to 7 amber, 1 to 3 red, so a 7 (“Good”) shows amber.

Overall score. The overall score is the average of the two halves, rounded to the nearest whole number, with a half rounding up. It appears only when both halves have a score. Health and planet count equally. It uses the For you words and colours.

Same for everyone. Scores are not currently personalised. Everyone who scans a product sees the same assessment, whatever preferences they have set. [Update if this changes.]

8. How Alternatives are found and ranked

Once a scanned product has an assessment, the app may show up to five Alternatives. This is how:

• Candidates. We look for products in the same category that do the same job and are sold in your country (worked out from your connection, defaulting to the UK), ideally in a major supermarket. The scanned product’s assessment is the baseline to beat. Where candidates are not already in the Database, an AI model searches the web and the same product databases for them.
• Selection. Candidates should be likely to score better on at least one half without being clearly worse on the other. At present that judgement is made by the AI when it suggests candidates. Each suggestion comes with a one-line reason that names any trade-off.
• Assess each one. Every candidate goes through the same assessment, with the same criteria and weightings, as any other product (section 2).
• Rank. The app lists them by the average of their two scores, highest first, with any unscored ones last. Tapping one opens its full assessment.
• Store. Suggestions are stored by barcode and country and reused for later scans.

What this means. We do not currently apply a separate rule to check that an Alternative scores higher than the scanned product, or that it is available near you. An Alternative may score the same or lower, may be out of stock or discontinued, or may not be a perfect substitute. Your thumbs-up or thumbs-down and shopping list do not change what is suggested. No brand, retailer or partner can pay to be suggested or ranked higher. If we add affiliate links, they will be labelled and will not change which products are suggested or how they are ranked [confirm].

9. Limits, and what you should not rely on

• Opinion, not fact. Scores, bands, Flags and Alternatives are our methodology-based opinion. They involve judgements about what matters, how much it weighs and where thresholds sit. Reasonable experts may disagree, and other apps may score the same product differently.
• Inputs can be wrong. Scores are only as good as the data and AI research behind them (sections 2 to 4). Some criterion scores are inferred rather than evidenced, especially where information is missing.
• AI can err. The AI can misidentify a product, misread a source, rely on a wrong web page or miss something. Because assessments are stored and reused, an error can repeat until it is corrected.
• Estimates, not measurements. Planet criteria use averaged or modelled data and are not a life-cycle assessment of the specific item, supply chain or batch. Health criteria are not a clinical assessment of any person.
• Not advice. Nothing here is medical, nutritional or allergy advice. Do not rely on the app for allergies or intolerances. Always read the label on the product you are buying.

Scores can change as data, science or this Methodology change, and are most useful for comparing similar products. They are not certifications or endorsements. Our Terms contain the full disclaimers and limits on liability.

10. Changes, corrections and contact

Changes. We may change this Methodology at any time to reflect new evidence, better data or product changes. We keep a changelog at [URL]. When the way scores are calculated changes materially, stored assessments are refreshed as needed.

Independence. No brand, retailer, advertiser, sponsor or affiliate partner can pay for, or otherwise influence, a score, a Flag or an Alternative. We do not manually adjust an individual product’s score for commercial reasons.

Corrections. If you think a score, Flag or product detail is wrong, tell us at hello@heyconscious.com. Brands and manufacturers can send a correction request with the barcode and supporting evidence, such as a current label. We correct factual input data. We do not change a score because a brand asks, and we do not accept payment in connection with corrections. See section 14 of our Terms.

Contact. Hey Conscious Ltd. Email hello@heyconscious.com.