Begin with a question that can be answered
Good research starts by defining its object. We specify the population or organism, the comparison, the outcome and the timescale. A question about a cellular process needs a different kind of answer from a question about human healthspan.
Before collecting evidence, we ask what would support the idea, what would weaken it, and what alternative explanation deserves serious attention.
The aim is a conclusion whose scope matches its evidence, including a clear account of what remains unknown.
Follow the claim back to its source
Reviews help us map a field. Claims that carry the argument need closer examination of original studies, methods, supplementary materials and relevant corrections.
We record what was actually read: an abstract, specific sections or the full text. We make gaps visible. Several papers from one trial are not several independent replications.
Where numbers matter, we check the outcome, denominator, comparison and uncertainty. Where we reproduce a calculation, we retain its inputs and assumptions. A calculation that runs correctly still needs sound inputs and an appropriate model.
Keep different kinds of evidence distinct
Mechanistic plausibility, a change in a biomarker and a demonstrated human benefit answer different questions. Our articles should explain the steps between them and the assumptions each step requires.
| Question | What the article should make clear |
|---|---|
| How might it work? | The mechanism, model and conditions under which it was studied. |
| What happened? | The measured outcome, comparison, population and time period. |
| How confident can we be? | Limitations, alternative explanations and the reason for the judgment. |
| Where does it apply? | The people or systems studied, safety information and the limits of extrapolation. |
We consider the strongest competing explanation. When the available evidence cannot distinguish between explanations, that uncertainty belongs in the conclusion.
Make the reasoning inspectable
Each research article should offer a readable account of its central question, the evidence supporting its main claims, the reasoning connecting them and the important limits. Sources are linked so readers can look further.
Authorship and review describe work that actually happened. AI-assisted work is identified as such. A second reading by the same author is a self-review; review by another AI agent is not a clinical expert review.
We normally publish explanations of evidence. Actionable advice about medication, dose changes or special populations requires appropriate human professional review within that scope.
Revise as understanding changes
Articles will show an evidence cutoff and an update history. A new study, a correction to a source or a flaw in our reasoning can trigger a revision. Changes that affect a conclusion need an explanation.
Chinese and English articles share the same underlying evidence. Each language is checked for its meaning, including comparisons, units, uncertainty and qualifications. A translation becomes outdated when the substance of its source changes.
The research collection has not yet opened. These are the standards we will use to develop and review its first articles.