How ingredient scores are calculated

Open any scanner app and you get a number. Green and 82. Red and 2. A letter. A face.

That number is doing an enormous amount of work, and almost nobody is told how it was built. This page is about the general shape of it, because once you can see the machinery you can judge for yourself when to trust the output.

The rough anatomy of a score

Most scoring systems, whatever the app, combine some subset of these.

An ingredient-level hazard rating. Each ingredient is assigned a concern level from published toxicology, regulatory classifications and the research literature. This is the biggest input in cosmetics-oriented apps.

A nutritional component. In food apps, some established measure of nutritional quality, often derived from public-health work such as Nutri-Score, plus the degree of processing, for which the NOVA classification is a common reference.

Weighting. Not every ingredient counts equally. Position in the list is a proxy for quantity, so things near the front usually weigh more.

Penalties and bonuses. Points off for specific categories of concern, points on for things like organic certification.

A data-gap rule. What to do when nobody knows. Some systems treat missing data as a reason to score worse, on precautionary grounds.

Then it is normalised onto a scale people can read at a glance.

What a score carries well

Comparison. Standing in an aisle holding two jars, a number answers “which of these two” faster than any sentence, and that is a genuine, daily, useful thing.

Consistency. The same product gets the same answer every time, and two people can talk about it.

Attention. Scores got millions of people to turn packages over, which is a real public good regardless of what you think of the methodology.

What a score structurally cannot carry

Exposure. A hazard-weighted number describes what ingredients can do, not how much reaches you. That distinction is the whole of hazard versus exposure, and it is the most common criticism made of hazard-led scoring by cosmetic chemists and toxicologists.

You. This is the one we care most about. A score has to mean the same thing for everybody, which means it cannot know about your sesame allergy, your eczema, or the fact that you are reading labels differently for nine months. A product can score well and be exactly the thing you must not buy.

Uncertainty. A number is equally confident about an ingredient with fifty years of study and one with almost none. Some systems penalise the second, which sounds cautious and can mean a well-characterised newer molecule scores worse than an older one purely because less has been written about it.

Formulation. Ingredients are usually rated in isolation, but products are formulations. Concentration, pH, what else is in there and what it is used for all change the answer, and the score generally cannot see any of it. This is why Retinol is not one thing and Potassium sorbate at preservative level is not the same as at any other level.

Trade-offs. Removing a preservative does not give you a product with no preservative. It gives you a different one, or a shorter shelf life. Scores rarely price the alternative.

So why does Pura not have one

Because we could not work out how to build a number that answers the question people are actually asking.

“Is this good” is not a property of a product. “Is this a problem for me” is a real question with a real answer, and it changes per person, which means it cannot be a shared score. So instead of a number, Pura tells you what is in the thing, flags what matches your own profile, and lets you ask why in plain words.

We should be honest about the cost. Our answer is slower. It is harder to compare two products at a glance. It does not screenshot well, and it will never spread the way a red 12 out of 100 spreads. Some people will find that worse, and for the aisle-comparison use case they are right.

How to use a score well

Treat it as a prompt rather than a verdict. It is good at telling you which product is worth a closer look and bad at telling you what to do.

Then ask the three questions that a number cannot answer for you: at what dose, by what route, and does any of this involve something on my own list. The rest of the method is in how to read an ingredient label, and the specific arguments about individual apps are on the EWG page and the Yuka page.