1. Overview
nemi Score aggregates each product's ingredient profile against regulatory bodies, toxicology authorities, and clinical research databases from around the world, using a deterministic algorithm. AI does not assign scores. (AI is used to generate ingredient explanations — see our AI Policy for details.)
This page documents the inputs and structural framework of our scoring system at the level needed to evaluate its independence and methodological credibility. The exact weights, thresholds, and computational rules are proprietary and not disclosed, to prevent replication by competitors.
A score does not guarantee or deny that any specific product is "safe" or "dangerous." It is a reference indicator derived from publicly available toxicology data mapped at the ingredient level. See §9 (Limitations and disclaimer) for further detail.
2. Data sources
We avoid dependence on any single database. The system synthesizes evidence from multiple independent sources and cross-validates them. All inputs are publicly available.
2.1 National and regional cosmetic regulations
- Japan: 化粧品基準 (Ministerial Notification No. 331 of 2000) — list of prohibited and restricted cosmetic ingredients
- Japan: MHLW Quasi-Drug Active Ingredient Lists (薬食審査発 notifications relating to medicated cosmetics)
- EU Cosmetics Regulation 1223/2009 — Annex II (banned substances), Annex III (restricted substances), Annex IV (colorants), Annex V (preservatives), Annex VI (UV filters), through the latest amendments (including 2023/1545)
- SCCS (Scientific Committee on Consumer Safety) Opinions — European Commission ingredient-level safety evaluations published as PDFs
- US FDA — cosmetics warnings, recalls, and Color Additive regulations
- Health Canada Cosmetic Ingredient Hotlist — prohibited and restricted ingredient list
- AICIS (Australian Industrial Chemicals Introduction Scheme; formerly NICNAS)
- ASEAN Cosmetic Directive — regional banned-substance list
- China NMPA: IECIC (Inventory of Existing Cosmetic Ingredients in China)
- Korea MFDS (Ministry of Food and Drug Safety) cosmetic ingredient prohibitions and restrictions
2.2 International toxicology and chemical-safety bodies
- IARC Monographs (International Agency for Research on Cancer) — carcinogenicity classification (Group 1, 2A, 2B, 3)
- ECHA (European Chemicals Agency) — REACH/CLP classification and registration
- WHO IPCS Environmental Health Criteria — International Programme on Chemical Safety assessments
- OECD Test Guidelines — standard methods for skin irritation, sensitization, and other endpoints
- CIR (Cosmetic Ingredient Review) — Final Reports of the U.S. expert panel that reviews cosmetic ingredient safety
2.3 Industry registers and naming standards
- JCIA INCI register (Japan Cosmetic Industry Association) — official Japanese display name to INCI mappings
- IFRA Standards (International Fragrance Association) — fragrance-ingredient use limits and amendments
- INCI Common Ingredients Glossary — international cosmetic-ingredient naming convention
2.4 Clinical surveillance and peer-reviewed literature
- NACDG (North American Contact Dermatitis Group) baseline series surveillance
- IVDK (Information Network of Departments of Dermatology, Germany) patch-test statistics
- ESSCA (European Surveillance System on Contact Allergies) aggregated data
- JSCD (Japanese Society for Contact Dermatitis) Japanese baseline series
- Peer-reviewed journals: Contact Dermatitis, Journal of the American Academy of Dermatology, British Journal of Dermatology, Journal of Cosmetic Dermatology, International Journal of Cosmetic Science, and similar
2.5 Other hazard databases
- EWG Skin Deep (Environmental Working Group, US) — consumer-facing hazard scoring
- cosmetic-ingredients.org — per-ingredient clinical-trial data (skin irritation, sensitization, etc.)
3. Methodological framework
Scores are computed via a multi-layered evaluation architecture. The individual formulas are not disclosed, but the overall structure is the following.
3.1 Ingredient identification and naming normalization
Japanese label names are resolved to international INCI names using the JCIA register, the INCI Common Ingredients Glossary, and historical synonym data. Spelling variations, partial matches, and deprecated old names all collapse into a canonical, single ingredient identifier.
3.2 Cross-validation across sources
For each ingredient, all sources from §2 are queried independently to retrieve regulatory status, toxicology classification, and clinical findings. Where sources disagree, we apply the weight-of-evidence principle: more recent, larger, and methodologically stronger studies are prioritized.
3.3 Exposure-aware evaluation
Rather than a binary "harmful / harmless" judgment, the framework incorporates use-pattern aware exposure assessment. Rinse-off products (cleansers, shampoos, body washes), leave-on products (toners, moisturizers, sunscreens), and products applied near mucosa (eye makeup, lip products, oral-care) carry materially different real-world exposures for the same ingredient. Our framework reflects the Retention Factor methodology described in the SCCS Notes of Guidance (latest revision).
3.4 Low-dose-reactive ingredient handling
Some ingredients elicit reactions at trace concentrations once a person is sensitized — a clinical reality that simple concentration-dependent scoring does not capture. EU Annex III fragrance allergens, certain preservatives, formaldehyde donors, and similar compounds are flagged individually and handled separately from standard concentration-based logic.
3.5 Disclosure-status adjustment
Some ingredients (Parfum / fragrance, ± colorants) are legally exempt from composition disclosure under Japanese law and most other regulatory frameworks. Individual safety evaluation for these is structurally impossible. Products containing them receive a separate transparency adjustment, and the situation is surfaced explicitly on the product detail screen.
3.6 Deterministic aggregation
The final aggregation step applies the same rules mechanically across every product. There is no AI-driven subjective judgment, and no per-product exception handling. Two products with identical ingredient composition and identical use pattern always receive identical scores.
4. Evidence-quality grading
Not every ingredient has the same depth of research support. We grade evidence quality along the following dimensions:
- Methodological quality: human clinical > animal > in vitro, weighted accordingly
- Study scale and reproducibility: claims from a single study are held provisional until replicated or confirmed by meta-analysis
- Independent regulatory review: if any of the bodies in §2.1–2.2 has independently evaluated the ingredient, that evaluation is prioritized
- Missing data: when no data is found, the ingredient is NOT treated as "dangerous." It is flagged as "currently not evaluable"
5. Continuous monitoring
Cosmetic regulation is updated frequently worldwide. We refresh methodology and data on the following cadence:
- Continuous monitoring of EU Omnibus amendments, SCCS opinions, and CIR Final Reports as they publish
- Re-evaluation of affected ingredients when IARC Monographs add or revise entries
- National regulatory body updates (Japan, US, EU, Canada, Australia, ASEAN, China, Korea) reflected as they happen
- Annual integration of clinical surveillance data (NACDG, IVDK, ESSCA, JSCD)
6. Cases that cannot be scored
We either suppress the numeric score or flag a caveat in the following cases:
- Quasi-drugs (医薬部外品): Japanese medicated cosmetics. Full ingredient disclosure is not legally required, and remaining ingredients are not required to be listed in concentration order. Numeric scoring is not currently applied to this category.
- Novel or rare ingredients: ingredients not yet present in standard research databases, with limited published safety data. Handled conservatively and flagged.
- Products with high non-disclosed-ingredient ratio: when fragrance or other non-disclosed components form a large share of the formulation, the reduced reliability is surfaced explicitly.
7. How "similar products" are selected
The "similar products" section on a product page is generated by these criteria:
- Candidates are drawn from the same product category (cleanser → cleanser, toner → toner).
- Ranking is by ingredient profile similarity, not by score.
- Each similar product is shown with its own independently-computed score. Users make the comparison themselves.
- Affiliate retailers, commission rates, and sponsorship status do NOT influence ranking or selection in any way. See affiliate disclosure for details.
The similar-products feature is a neutral discovery tool for finding other options in the same category. It does not represent nemi's verdict on the relative safety of one product versus another.
8. Separation from affiliate revenue
Buttons on the product detail page like "View on Amazon", "Buy on Rakuten", "View on Qoo10", and "View on @cosme" are affiliate links to partner retail sites. If a user makes a purchase via these links, nemi receives a commission from the partner.
This affiliate relationship has no impact on product scores, ingredient safety evaluation, or the similar-products selection algorithm. The scoring logic is, by design, blind to commercial relationships. Full disclosure at affiliate disclosure.
9. Limitations and disclaimer
- Scores are a snapshot of publicly available safety data at a given moment. When brands reformulate, there can be a lag before our database catches up.
- This is not medical advice. Scores do not substitute for dermatology, pharmacy, or medical professional advice. For skin issues, pregnancy, breastfeeding, known allergies, or atopic dermatitis, please consult a medical professional.
- Scores do not guarantee or deny the quality, safety, or efficacy of any specific product. They are independent reference information based on published scientific data.
- Skin condition and sensitivity vary between individuals. An ingredient considered safe for the general population may still elicit a reaction in a specific person.
10. Changelog
- May 9, 2026: Methodology section comprehensively rewritten; reference-source list expanded
- May 5, 2026: Initial publication
When the scoring algorithm is updated, we record the date and the nature of the change here so we can transparently explain to brands or users why a score moved over time.
11. Reporting errors and inquiries
If you find an error in a score or ingredient, or have questions or feedback about our methodology, please use the in-app "Report issue" flow, or email support@nemi.beauty. Brand operators may also request re-evaluation; we respond on the basis of publicly available evidence.