Evidence methodology
How the research works
From a search lead to a published claim: the full path, and every point where a claim can be stopped.
The research-to-publication pipeline
- Research lead — a candidate source identified by search or a known publication.
- Source verification — the URL is retrieved. Only then does a record become FETCHED.
- Evidence assessment — the study type, relevance and strength are assessed and rated.
- Claim extraction — the specific assertion is written as a standalone claim record.
- Content draft — a draft links to claims; claims link to sources.
- Editorial review — a human editor checks the drafting and the sourcing.
- Medical-safety review — where applicable, a qualified reviewer assesses safety-sensitive content.
- Publication — server-side gates run. Blocked content does not go live as evidence-backed.
- Repurposing — derived formats inherit the original claim and its limitations.
Provenance labels
These are the labels from the original research system, used verbatim.
- FETCHED
- The source URL was fetched and the value was extracted from the fetched content.
- SEARCH_SNIPPET
- The value came from a search-result excerpt; the underlying page was NOT fetched. A lead requiring verification before use in content.
- NOT VERIFIED
- Verification was attempted; the value could not be confirmed.
- NOT RESEARCHED
- Work remains outstanding.
Evidence ratings
Ratings describe the strength of the evidence for the thing being rated. They are never inferred — if a source has not been assessed, it reads as unassessed.
- —
- Not yet assessed
- ★★★★★
- Very strong evidence
- ★★★★
- Strong evidence
- ★★★
- Moderate evidence
- ★★
- Limited evidence
- ★
- Weak evidence
- ⚠️
- Marketing claim / insufficient evidence
Evidence categories
- SCIENTIFIC EVIDENCE
- Peer-reviewed research or systematic reviews.
- CLINICAL/DERMATOLOGICAL CONSENSUS
- Established dermatological practice consensus.
- CONSUMER EXPERIENCE
- Reported consumer experience — not clinical evidence.
- SOCIAL-MEDIA POPULARITY
- Popularity signal only — says nothing about efficacy.
- MARKETING CLAIM
- A claim made by a brand — never present as science.
- OWN INFERENCE
- Explicitly labelled inference by the editorial/research team.
What counts as a fetched source
A source is FETCHED when the underlying page was retrieved during the research session and the value recorded was extracted from the retrieved content — not from a search-engine excerpt, and not from a summary of a summary. Where only an abstract was retrieved, we record that the full text was not extracted, because the algorithms in a full paper and the conclusions in its abstract are not equivalent.
Known limitations of this checkpoint
The research this platform is built on is a work-in-progress checkpoint, and the gaps are material:
- Five of eighteen clinical evidence records were fetched; the remainder are search-snippet leads.
- All fifteen African brand records are snippet-level and unverified.
- Influencer demographics are not verified, and creator metrics are not collected.
- Country-specific search demand was not collected beyond one retrieved trends paper.
- The viral content database was not collected at all.
- Five of nine hundred and fifty planned content ideas exist.
We publish this inventory because a reader who knows the size of the gap can weigh our output appropriately. See the research library for the record-level view.
Re-verification
Prices are re-checked monthly. Affiliate terms are re-checked quarterly. Search-snippet sources are upgraded to FETCHED only when the underlying page is actually retrieved — never the other way around.