Evidence grading
Every conclusion says where it came from
What slows product people down is not a shortage of data, it is not knowing which line was measured and which line a model guessed. Every conclusion we publish carries an A–E tag: who produced it, where it came from, whether it still needs checking. You never have to ask.
The five levels
A · Verified in-house
Made and measured by us: sampling, stability, challenge tests, pack compatibility, production batches, customer feedback.
We use it for Numbers we produced ourselves: sampling, stability and challenge tests, pack compatibility, production batches, customer feedback. Shown inside your project, not spread across public pages.
B · Verificado por terceros
Verified by a supplier or third party: raw-material technical data, lab reports, university work.
We use it for Raw-material technical data, third-party lab reports, university work. Originals are shown when your project reaches formulation and compliance.
C · Fuente oficial
Regulations and authoritative public sources: official rules, standards, official inventories, trade bodies, literature.
We use it for The market regulation pages, ingredient inventories (NMPA names, CosIng functions, national annex restrictions), and the name, function and restriction rating of each ingredient on a decoder page. Each carries its official source.
D · Market observation
Market observation: e-commerce, social, reviews, new launches, prices, channels — with sample size and date.
We use it for Market research pages and the opportunity cards on the home page: bestseller boards, price bands, channels, launches and reviews. Figures carry sample size and date; without a sample size we do not publish them.
E · Inferencia de IA · por verificar
AI inference, not yet checked by a person or a test. Treat it as a candidate, not a conclusion.
We use it for Worth-making verdicts, AI-written formulation views, and the category and packaging an AI read off a product photo. Always marked to be verified: candidates, not conclusions.
Three hard rules
- E never gets promoted to A on its own.Re-running a model a hundred times leaves it at E. Only a person moves a level, against new evidence: an official source takes it to C, a third-party report to B, and our own samples and measurements to A.
- A row is graded by its weakest part.An ingredient note may take its name and restrictions from official inventories (C) while the line on what it does in a formula is model-written. The whole row is then marked E: C never lends its halo to E.
- No figure goes out without a sample size.Market observations state how many products, which days, which board. If we cannot state the sample, we drop the number rather than dress it up in adjectives.
Where you see it
- Ingredient decoder: every note and verdict tagged →
- Market regulations: 42 markets, each with its official source →
- Global market research: with sample sizes and dates →
- Product library: AI-read details marked to be verified →
Want harder evidence behind something marked E? Tell us the product, and inside a project we take it up level by level, C to B to A. Start a product project →