About · Finuwell

About this site

Method

• · Every number and mechanism claim has at least one source (DRI / EFSA / review / textbook).
• · Every citation carries two things: the study design (guideline, review, randomized trial, cohort, mechanistic study…) and the certainty of the evidence, on one four-step scale used across the site: high (several consistent randomized trials or a systematic review of them, or a strong guideline recommendation) · moderate (one large randomized trial, trials that do not fully agree, or consistent large cohorts) · low (mostly observational studies, or small, short or surrogate-endpoint trials) · very low (case reports, animal or cell studies, mechanistic reasoning). Expert-opinion sources are marked “consensus.” The prose no longer uses A / B / C letters.
• · The health report runs on an explicit rule engine — no AI model infers anything about your input.
• · Health-report input stays in your browser by default. It is uploaded — encrypted, to your account — only if you tick “save this report”.
• · The assistant is the exception: your question is sent to a third-party model provider to generate an answer. The privacy policy names every provider and what it handles.

What we don't do

• · No disease diagnosis or prescription advice.
• · No brand recommendations, no ads, no embedded shopping links.
• · No advertising profiles; we never sell or trade your data.
• · No nagging, no popups, no FOMO marketing.

Who writes this, and how we keep it right

This site is written and maintained by an independent developer who is not a practising clinician. Which is exactly why it does not ask you to trust the author. It runs on a process you can check — every line below is enforced automatically, not promised:

• · No claim without a source: every scene carries at least one registered citation, or the build fails.
• · Sources must resolve: every record with a DOI is checked against Crossref. That is how a fabricated reference — wrong author, journal, volume and DOI — was caught.
• · Sources must fit the claim: three further audits hunt for citations that are real but unrelated — years that cannot line up, subjects that do not overlap, studies named in the prose but never attached.
• · Numbers get re-read: the corpus is re-checked against the original papers in batches; confirmed corrections are recorded in a public work log.
• · Every article lists its references with clickable DOIs — you do not have to take our word for it.

None of this replaces clinical judgement, and none of it catches the hardest error — a real paper that simply does not say the thing being claimed. That takes a person reading the original. If you find anything wrong, write to huahaoshang2000@gmail.com; we will correct it and note the change.