Accuracy is a process, not a tone of voice.
These rules govern how we select, explain, label, correct, and illustrate research.
Start with the primary source, then check government records
We link to the original paper, report or dataset whenever it is publicly available. Relevant federal indexes, trial registries, repositories, datasets and agency guidance are checked for verification and context. Their role is labeled because a registry, repository and guidance page do not provide the same kind of evidence.
Match the verb to the method
Observational studies can find associations. Controlled experiments can support causal conclusions under defined conditions. Models generate estimates. We do not blur those distinctions for a stronger headline.
Say what was actually studied
Human, animal, cell, material, and simulation results are labeled clearly. We avoid implying that a result in one setting has already been demonstrated in another.
Keep uncertainty and negative space
Limitations, uncertainty intervals, plausible alternative explanations, corrections, and conflicting evidence belong in the main explanation—not buried after the conclusion.
Use facts; do not copy expression
Our summaries are original explanations based on research and cited sources. We do not reproduce abstracts, articles, press releases, charts or photographs. Material on a government website is checked for third-party rights, attribution requirements, trademarks and image restrictions before any reuse.
Correct transparently
Material factual errors are corrected promptly. When a correction changes the interpretation, evidence grade, or conclusion, the article receives a visible update note.
Separate editorial work from advertising
Advertising and future sponsored material will be labeled and will not purchase favorable conclusions, evidence grades, or article placement presented as editorial judgment.
Publish only after human review
Our workflow is Discover → Locate Primary Research → Verify → Classify → Explain → Cite → Review → Publish. AI and feeds may assist discovery, but nothing becomes public automatically.
Check for duplicates and changes
We compare DOI, PMID, title, journal and authors before adding a record. Corrected, retracted, superseded and expression-of-concern records remain visibly labeled and do not contribute normally to evidence summaries.
Use only verified imagery
We use original work, verified public-domain material, properly licensed Creative Commons images or explicit permission. If reuse rights cannot be verified, the image is not used.