How we grade evidence
Every claim on this site carries a grade. The grade says what kind of evidence stands behind the sentence, so a confident sentence never stands in for strong evidence.
| Grade | What it means | Where it comes from | In plain words |
|---|---|---|---|
| E1Strong | Randomized trials in people, or reviews that pool them | Randomized trial, meta-analysis, systematic review | Chance decides who gets which treatment, so differences in results come from the treatment. |
| E2Good | Careful observation in people, or a guideline built on the evidence | Observational study, guideline, practice parameter, consensus statement | Real people, carefully studied, or experts' recommendations drawn from the evidence. Strong, but not a trial. |
| E3Early | Laboratory or animal studies only | Laboratory and animal research | Shows how something might work. Not yet shown in people. |
| E4A link, not proof | An association; cause is not shown | Studies that find two things occur together | A link, not proof. Whether one causes the other is unknown. |
| E5An idea | Proposed, not yet tested | Hypotheses | An idea worth testing. |
| Official text | An FDA label or the law, quoted as written | FDA prescribing information, Drug Facts, federal law | Reported as written rather than graded. Labels set out what a medicine is approved for and its warnings. |
| Review | An expert review; not graded | Expert reviews and articles | Useful summaries by experts. They are not given a grade of their own. |
The rules
- The grade comes from the source. Each source in the library is labeled with its type (randomized trial, guideline, FDA label and so on), and the type sets the grade. Para AI in the Epitope app uses the same table.
- A sentence shows its strongest source. When a sentence rests on more than one source, it shows the strongest grade among them.
- Labels and the law are quoted, not graded. An FDA label is what the FDA approved. It is marked "Official text" and reported as written. Dose tables from labels are shown whole, never as one row picked for one person.
- Expert reviews are not graded. They are marked "Review" so that a summary is never mistaken for a trial.
- A grade is about the evidence, not about you. Strong evidence that a treatment works in a group does not decide whether it fits one person. That decision belongs with a clinician who knows the person.
How sources are checked
Every guide is written only from excerpts in the verified libraries. Each excerpt is copied word for word from its source: the PubMed abstract, the DailyMed label, or the original page. A check script fetches every source again and fails if any excerpt no longer appears word for word. Primary studies must also meet a citation bar (100 or more citations in NIH iCite, or a relative citation ratio of 3 or more for papers from 2021 on); guidelines, consensus statements and labels are exempt.
Under each guide's sources you can open "The words this page rests on" and read the exact excerpts.
The libraries today
| Library | Sources | Version |
|---|---|---|
| Allergy library | 85 | October 16, 2026 |
| Surgery library | 69 | October 18, 2026 |
| Website supplement | 32 | October 18, 2026 |
The allergy and surgery libraries are the same ones Para AI answers from. The website supplement adds sources for topics the app's libraries do not yet cover (environmental allergy and mold, IgG panels, immunotherapy for pollen and dust mite, eczema, and the procedure pages), checked by the same method. See what changed in the library.
What we will not do
We will not grade a claim higher than its evidence, add a fact a source does not state, or name a clinic, product or clinician as the right one. Where the evidence cannot answer a question yet, the guide says so.