About — Healthspanner

Healthspanner is a reading library for healthy adults — what actually works for healthspan, what doesn't, and how strong the evidence is. Distilled from peer-reviewed literature, graded by evidence strength, and rewritten for clarity.

The source corpus

The site is built on top of roughly two dozen long-form research reviews — each one a 40–80 KB synthesis of a single domain (sleep, VO₂ max, protein and mTOR, sauna, alcohol, geroprotectors, and so on). Those reviews in turn cite thousands of primary sources: randomized controlled trials, Mendelian-randomization analyses, prospective cohorts, meta-analyses, and the occasional landmark mechanistic paper.

Wherever a claim on this site is meaningful enough to anchor a recommendation, you should be able to find its source named in the running text — by trial, journal, and year (e.g. “PREDIMED, NEJM 2018”) — and a hyperlink to the paper or trial registry where one exists.

The method: AI-assisted deep research

Each of the underlying research reviews was produced using agentic deep-research models that read and cross-checked hundreds of papers per topic, surfaced the load-bearing citations, and reconciled conflicting findings. A human editor then distilled those reviews into the hierarchical structure you see here — homepage, pillar overview, deep dive — and tightened the prose for an average reader.

AI is good at breadth and recall; it is less reliable at calibration. So the editorial pass focuses on the things that actually matter: that effect sizes are quantified, that the study design behind each claim is named, that observational associations are not dressed up as causation, and that commercial bias (industry-funded trials, surrogate endpoints, short follow-up) is flagged.

Evidence grading

Every substantive health claim carries one of four tags:

Strong — multiple large RCTs or meta-analyses with hard endpoints (mortality, MI, dementia diagnosis).
Moderate — consistent RCTs but with heterogeneity, surrogate endpoints, or limited populations.
Weak / preliminary — small or single trials, mostly animal data, or only observational associations.
Caution — documented harm signal at common doses.

The grade lives next to the claim, not at the bottom of the page. A pillar overview tells you the rating up front; the deep-dive article behind it walks through the underlying studies.

Editorial principles

Plain language at the top, depth deeper down. Teasers and pillar pages are written for an average reader; acronyms and trial names appear once the context is set.
Quantify when possible.“30% reduction in CVD events” beats “significantly reduces.”
Acknowledge uncertainty. Observational data, Mendelian-randomization caveats, and industry funding are called out rather than hidden.
Don't moralize. Especially around alcohol, weight, or behavioral choices. The evidence is presented; readers decide.
No supplement, brand, or program promotion. Generic recommendations only.

What this site is not

Not medical advice. Nothing here is a prescription. Discuss any meaningful change with a clinician who knows your history.
Not exhaustive. The deep-dive articles are distillations. If a claim matters to you, follow the citation through to the primary source.
Not static. The evidence base moves — the alcohol J-curve has largely collapsed since 2019, the dementia-risk attributable fraction keeps climbing as new modifiers are identified. Pages are updated when meaningful new evidence lands.

Corrections and feedback