healthy100

Back to the list

How this was made and checked

Every item on the list went through the same steps. This page says what they were, and where they stop.

Research

The list is split into four domains: sleep and movement, food and substances, mind and social life, prevention and environment. GPT-6 Astra, an OpenAI model, researched each domain from public sources (guidelines, systematic reviews and large studies) against a fixed specification, and later wrote the practical steps the same way. Every item needs at least one Tier 1 source and one number a reader can find at that source. Every cited address was opened and the relevant text read; search snippets alone did not count.

Tier 1 is WHO, CDC, NIH, NHS, USPSTF, ECDC, a national guideline or Cochrane. Tier 2 is a peer reviewed meta-analysis or large cohort. Tier 3 is a reputable secondary source such as Harvard Health, and it never stands alone. Every item carries at least one Tier 1 source.

Fact-checking

Separate Claude Opus threads, Anthropic models working apart from the researcher, then fact-checked every item, one thread per domain, starting from the assumption that errors exist. They opened every link again, compared each number and its wording with what the source says, and flagged anything overstated or unsafe. Their corrections were applied before publication, and the practical steps went through the same check.

Every source on this site is a link. No claim here is meant to be taken on trust: open the source, read the number, and if we got it wrong the source will say so. That is the whole point of building it this way.

Merging and spot checks

An orchestrating AI agent merged the checked domains into one list, ran a validator that enforces the specification on every item, and spot-checked items against their sources before publishing. Pawel Jozefiak publishes the site and is accountable for it.

Clinical review

None. No doctor, nurse, dietitian or other clinician has reviewed this site or any item on it. The checks above are made by AI models against published sources, and they do not replace clinical review.

Translations

The other languages are translated from the checked English by Claude Opus. Only the prose is translated: numbers, sources and rules are copied, and the build puts any translated sentence whose numbers differ from the English back into English. A language is published once its interface is fully translated; until then its pages show the English text and ask search engines not to index them.

Evidence grades

The ranking

Every item gets a score from 0 to 100 for population level impact. It combines how much the thing changes death and disease with how common the gap is. A change with a huge effect that almost nobody needs scores lower than a moderate change that almost everybody needs. The general list is that score, in order.

The impact score is an editorial prioritization heuristic for the general list, not a validated estimate of your personal risk.

You give a few facts. Each item can carry rules: a hard filter that decides whether it applies to you at all (sometimes a list of alternatives, where any one of them is enough), and boosts that add or subtract points. Your score for an item is its impact score plus every boost that fires. Nothing is hidden: a field you skip fires no rule, and a rule you never see cannot move anything.

Tailoring uses rules written into each item. Every rule is listed on the home page, exactly as the engine reads it.

What this cannot promise

A second model checking a first is automated verification, not independent clinical review. No doctor wrote or approved these pages, and two models can misread a study the same way. The numbers come from studies of populations: they describe groups, not you. Where the evidence is observational the wording says the habit is linked to an outcome, not that it causes it, and linked effects must not be added together.

This is not medical advice. It is a reading list with numbers attached. It cannot know your history, your medication or your risk, and a few items carry a note about when to talk to a clinician first. If something here contradicts what your doctor told you about your own body, your doctor has more information than this website does.

Corrections

If a number, a source or a caution is wrong, email meet@jock.pl or message @joozio on X with the page address and what the source actually says. Corrections go into the data every page is built from.