How we score every sunscreen.
A composite 0–100 number summarises the formula across six axes — three deterministic (computed from the formulation), three judgement-driven (graded by an AI judgement model with a structured reasoning rubric). The composite is a weighted blend; the breakdown is always shown alongside it.
Every score on this site is a number you could recompute yourself. We grade each sunscreen on six axes, blend them with one published set of weights, and never adjust an individual product by hand — three axes are computed straight from the formulation, three are judged by an AI model against a fixed rubric, and the breakdown is always shown beside the composite.
The six axes
Segment width = each axis's share of the composite; the six weights sum to 100%.
- UV protection 25%
How well the product blocks UVA + UVB across the spectrum at the stated SPF / PA.
Deterministic - Health safety 25%
Endocrine concerns, systemic absorption signals across the formula.
AI-judged - Sensitivity safety 15%
Risk for reactive skin — fragrance, alcohol, known sensitisers.
AI-judged - Cosmetic elegance 15%
Spread, finish, white-cast, comfort on skin.
AI-judged - Filter modernity 10%
Penalises legacy filters (octinoxate, oxybenzone); rewards photostable next-gen filters.
Deterministic - Environmental safety 10%
Reef toxicity, biodegradability of UV filters and surfactants.
Deterministic
What "composite" means
Composite is a deterministic weighted sum:
composite = 0.25·uv + 0.10·modernity + 0.15·sensitivity + 0.10·environment + 0.15·elegance + 0.25·health
Weights are pinned in our scoring config; any change bumps the eval version so every product re-scores in lockstep. We never tweak a single product's score by hand.
Deterministic vs AI-judged
Deterministic axes are computed from structured fields in our product database and the joined UV-filter properties in our ingredient database — no model in the loop, fully reproducible. Judgement axes are scored by an AI judgement model with a structured-output rubric, with the reasoning summary stored alongside the scores so you can read why a product got the number it got.
Computed from structured database fields and joined UV-filter properties. No model in the loop, fully reproducible.
45%of the composite
Graded by an AI judgement model against a pinned, versioned rubric, with the reasoning stored alongside the score.
55%of the composite
Scores come from the pipeline, not the sales team. No brand can pay to rank, and no product score is ever edited by hand.
What we don't do
- No paid placement. Brands and retailers cannot pay to be in or move up the rankings.
- No editorial overrides. Scores are produced by the pipeline; we don't bump a favourite up or a disliked product down.
- No invented data. Where a field is unknown for a product, the page shows it as unknown rather than guessing.
- No re-scoring on retailer change. Rankings are computed before commercial terms; new retailer listings don't shift scores.
Frequently asked questions
Can brands pay for placement in the rankings?
No. Brands and retailers cannot pay to be included, excluded, or moved. Scores are produced by a pipeline from formulation data; affiliate commercial terms are applied after scoring and have zero effect on rank.
How is the AI-judged part of the score kept honest?
The three judgement axes are scored against a pinned, versioned rubric with structured output — the model returns numbers plus a written rationale that is stored and displayed with the product. Any rubric change bumps the eval version and re-scores every product in lockstep, and regression tests pin anchor products to expected score bands. No individual product score is ever edited by hand.
Why did a product’s score change?
Two reasons only: the formulation changed (a new INCI list re-triggers scoring) or the eval version changed (every product re-scores together). New retailer listings, prices, or affiliate terms never shift a score.
Do affiliate links affect the rankings?
No. Buy buttons route through retailer affiliate programs and we earn a commission on qualifying purchases — that revenue keeps the catalogue free. Rankings are computed before commercial terms, and a product with no affiliate relationship ranks exactly as its formula scores. Disclosure appears on every page with affiliate links, in EN / JP / KR.
How fresh is the data?
The catalogue is refreshed regularly: availability is re-checked (products no longer purchasable are removed rather than hidden), and regulatory watchers monitor the JP / KR / EU / US allowed-filter lists. Every refresh lands as a reviewed change before it ships.
Where does the ingredient data come from?
INCI lists are scraped from retailer listings and resolved against regulatory sources: the FDA OTC monograph (US), MHLW quasi-drug list (Japan), MFDS list (Korea), and Annex VI of EC Regulation 1223/2009 (EU). Unresolvable names are queued for human review instead of being guessed.
Sources & references
- FDA OTC sunscreen monograph (United States)
- MHLW quasi-drug ingredient list (Japan)
- MFDS approved-filter list (Korea)
- Annex VI, EC Regulation 1223/2009 (EU)