kleerer.io

Source for kleerer.com — an independent, evidence-based health-data project. The site is static (no build step, no backend, no runtime dependencies) and is published via GitHub Pages directly from this repository.

What this is

kleerer. makes supplement science available to people looking for advice or recommendations, without sponsors, affiliate links, or paid placement. It's built in numbered "p" steps (products); the first one is live.

• index.html / style.css / script.js — the root landing page: the project's mission statement, bilingual (EN/FR, detected from the browser's language, never from IP), linking through to each product.
• compare/ — p1 · compare, the first product: an independent supplement comparator. Open compare/index.html in a browser or visit kleerer.com/compare. See compare/README.md for how it works and compare/METHODOLOGY.md for the full scoring rules.
• bot/ — a transparency page describing kleerer-bot, the crawler used to collect price/label data (identifies itself honestly, respects robots.txt, rate-limited, no login walls bypassed). Site owners can request exclusion via hello@kleerer.com.
• compare/evidence/ — what supplements actually do, by need. Sleep, stress, muscle, bone, heart, immunity, memory, joints, skin, gut, energy, libido — each supplement graded A–D on the trials behind that specific pairing, participants counted, every figure cited. The EFSA authorised claim is shown separately from the trial evidence, because a regulator approving a sentence about a nutrient's role is not a trial showing an outcome improved. The dataset is one readable file (evidence.js) so any grade can be challenged with a better study.
• terms/ — terms of use, including the database-right and text-and-data-mining clauses (EN/FR).
• NOTICE.md — which licence applies to which layer, and exactly what the database right does and does not claim. Read this before reusing anything.
• robots.txt / .well-known/tdmrep.json — the TDM reservation in human- and machine-readable form.
• analytics.js — cookieless audience measurement: no cookies, no identifiers, no personal data, DNT/GPC honoured, so no consent banner and no self-selected baseline. Inert until a provider is configured.
• manifest.webmanifest / sw.js / pwa.js / icons/ — the installable app. compare/ is a PWA: add it to a home screen and the whole catalogue works offline, which is the point — the moment you most want a score is standing at the shelf, which is exactly where the signal dies. The ~390 KB catalogue is precached, so it runs with the radio off. The service worker touches our own origin only, stores nothing about the visitor, and sends nothing anywhere.
• CNAME — GitHub Pages custom-domain config, points the repo at kleerer.com.

What it's used for

compare/ scores dietary supplements (whey, creatine, vitamin D3, magnesium, omega-3, multivitamins, zinc, vitamin C, collagen, probiotics, melatonin, vitamin B12, vitamin K2, B-complex, biotin, folate — and, shown apart, the botanicals ashwagandha, maca, rhodiola and curcumin, and in teal the testosterone boosters tribulus, fenugreek and ZMA) on a transparent, reproducible 0–100 Health & Compo Score (grades A–E), computed only from what's on the label and what a brand publicly proves — never from marketing claims:

• Health & Compo Score = Composition & Efficacy (0–40) + Purity & Additives (0–30) + Transparency & Testing (0–30). A banned/restricted/withdrawn substance (per EU-EFSA, US-FDA, or WHO — strictest of the three wins) is an automatic 0 / grade E, regardless of anything else.
• Price per gram of active ingredient, separated from the health score — price never affects the score, only the value comparison.
• Personalised dosage check against the user's own targets, including a stack builder (p2, beta) that sums combined products (built-in secondary actives included) and flags when a combined total crosses an EU upper limit.
• Everything is open-method: the auto-tagger and scoring engine are plain, auditable Python scripts, so anyone can challenge a score by pointing at better proof.
• An unreadable label is not a score. If the auto-tagger meets an ingredient string it cannot classify, the product is withheld from the site entirely until a human rules on that string — it is not published with a token penalty. Rulings are recorded in compare/data/review-ledger.json with the reason, reviewer and date, and each published product shows what was decided about it. See METHODOLOGY §2c and compare/scripts/review.py.

The data

• Botanicals are scored apart. Ashwagandha, maca, rhodiola and curcumin are plant extracts, not nutrients — no reference intake, no EFSA limit, no authorised claim — so they appear in light purple, go through the same score and review gate as everything else, and carry their safety advisories where the dose would otherwise be. See METHODOLOGY §5.
• Testosterone boosters are scored apart too, in teal. Tribulus, fenugreek and ZMA go through the same score, which is exactly why they land low: no tribulus dose has raised testosterone in men in a controlled trial, and the evidence page grades most of the wider group (D-aspartic acid, turkesterone, boron, DHEA) D. The colour exists so a reader sees that verdict before the price. See METHODOLOGY §5b.
• Scope: 239 coded products across 23 categories (16 nutrient categories + 4 botanicals + 3 testosterone boosters), EU market with a French-market focus; prices snapshot July 2026 for the original nutrient categories, September 2026 for the botanicals, the vitamin batch (B12, K2, B-complex, biotin, folate) and the boosters. All 239 are published; the review queue is empty. Anything withheld in future shows up in compare/data.js under meta.n_withheld_for_review.
• compare/data/products_raw.json: 37 curated EU products in a coded schema (id, category, brand, price, form, dosing, additives, certifications, transparency flags, source URL, confidence level).
• compare/data/fr/*.json: ~202 French-market products as free-text labels (botanicals.json holds the 30 ashwagandha, maca, rhodiola and curcumin products, vitamins.json the 33 B12, K2, B-complex, biotin and folate products, boosters.json the 13 tribulus, fenugreek and ZMA products — all transcribed from the brands' own pages on 6 September 2026); compare/scripts/autotag.py auto-tags these into the same coded schema (form quality tiers, additive classification) used for the curated set.
• compare/data.js: the generated, scored dataset the app actually reads — built from the two sources above by compare/scripts/build_scores.py. Generated; not edited by hand.
• compare/pipeline/: the data-collection pipeline — a polite crawler (honours robots.txt, rate-limited, honest User-Agent) that snapshots price + label per product daily, diffs each snapshot against the previous day to catch price moves and silent reformulations, and enriches by EAN via OpenFoodFacts open data where possible. The code is public; the accumulating historical archive (the actual time-series asset) is kept in a separate private repository (kleerer-data) — see compare/pipeline/README.md.
• Data honesty: prices are EU/FR list prices as snapshotted in July 2026 (promo-heavy brands are flagged per product since they routinely sell well below list); each product carries a confidence level (high/medium/low); scores come from labels and public documents, not independent lab testing.

Disclaimer

kleerer. does not provide medical advice. Consult a healthcare professional before supplementing, especially if pregnant, on medication, or managing a condition.

Licence

Not one licence — one per layer. A software licence and a dataset are different kinds of thing, and applying MIT to both (as this repo previously did) licensed away the dataset by accident. Full detail in NOTICE.md.

Layer	Licence
Site/app code, scoring engine, auto-tagger, pipeline	AGPL-3.0-only (LICENSE) — also available under a commercial licence
compare/METHODOLOGY.md	CC BY-NC-ND 4.0 (details)
compare/data.js, compare/data/	CC BY-NC-SA 4.0 + database right (details)
Historical price/composition archive	Not published. All rights reserved.

The AGPL's network clause applies: run a modified version as a service and you owe its users your source. If that does not work for your product, the commercial licence is the paid exemption — hello@kleerer.com. It covers the engine only and never the archive.

Database right and TDM

The catalogue and the archive are protected databases (Directive 96/9/EC; CPI art. L.341-1 ff.), and text-and-data-mining rights are reserved (Directive (EU) 2019/790 art. 4(3); CPI art. L.122-5-3) — see robots.txt and .well-known/tdmrep.json.

The limit, stated plainly: the database right protects the database, not the facts in it. A single price is a fact and is not protected — quote or republish any individual price or label freely. What is reserved is extraction of a substantial part. The public pages stay deliberately crawlable and citable, including by AI assistants; the reservation targets bulk mining of the archive, the methodology, and the catalogue-as-a-database.

Corrections and pull requests are welcome. Contributions are accepted under the licence of the layer they touch.