How we review
How OncoForge builds its pages, who reviews them, and how we keep them honest and current. Educational only; not medical advice.
1. The studies are the authority, not us
Every health statement on this site is meant to report what published studies and guidelines describe, with the citation attached. We never tell you something works, cures, prevents, or treats cancer. Where the evidence is early, conflicting, or only from lab or animal studies, we say so. The goal is for a patient, a doctor, and a scientist to read the same page and all trust what it says.
2. How a source-grounded page is built
Our research pages aren't hand-written opinion. For a given subject we search PubMed for the relevant guidelines, reviews, and studies, then an AI model reads them in small batches — using the free full text where it's available, otherwise the abstract, never outside knowledge — and extracts the facts into a strict, structured form (model type, outcome direction, effect sizes, limitations, mechanisms). Every claim must quote the exact sentence from a cited source that supports it, and a deterministic check confirms that quote really appears in that source; a claim whose support can't be found is flagged. A second, stronger AI pass fact-checks each drafted claim against the source text and removes or flags anything the source doesn't support. Then a human reviews and approves the draft before it publishes. Nothing goes live automatically.
3. Which sources outrank which
When sources disagree, we don't quietly pick one. A deterministic ranking decides authority: practice guidelines and meta-analyses outrank single randomized trials, which outrank observational studies, which outrank case reports; a retracted source ranks at the bottom. Contradictions are shown on the page, with the more-authoritative source named, so you can see the disagreement instead of having it hidden.
4. Treatment maps: every option, sourced and labeled
Each cancer has a Treatment Map that puts every option in one place, organized by clinical readiness. Established (standard-of-care) treatments are extracted only from authoritative clinical sources — practice guidelines, consensus statements, and meta-analyses — each carrying a verbatim quote we verify against its source, then human-approved before it shows; we reference, but never redistribute, license-restricted guidelines like NCCN. Investigational and adjunct options — every drug, supplement, or repurposed agent the literature studies for that cancer — are placed by how strong their evidence is (tested in people, lab/animal only, or merely named in the literature). Each compound's regulatory status comes from the U.S. FDA's open drug-label data, so it's honestly marked as FDA-approved for that cancer, used off-label, or not an approved drug at all. The map shows what has been studied and how strongly — never a recommendation that anything works.
5. Reported figures: numbers straight from the studies
Where a study reports a quantitative result — a survival time, a hazard ratio, a response or side-effect rate — we capture it as typed data and show it tied to the exact regimen, the comparison group, the patient population, and the number of people in the study, with its source one click away. We copy these numbers; we never average them, never combine figures from different studies, and never turn them into a prediction for any individual. We draw effect sizes as a forest plot — each result a point with its confidence interval on a log scale around the no-difference line — so the uncertainty is visible: a wide interval, or one crossing that line, reads as weak or not significant, and a result with no reported interval is shown as a hollow point. Every mark is computed only from these typed numbers, never from prose, and is labeled in text as well. Doses shown are those used in studies, not a dosing guide. The relevant clinical trials are tracked straight from ClinicalTrials.gov: their status and dates are the registry's own — a trial is shown as completed only when the registry says so, and completed is not the same as positive, so we link the results for you to read. Anything we can't trace to a cited source is dropped, not shown.
6. Kept current, with no drift
Published pages are re-checked against PubMed for retractions and updates. If a cited study is retracted, the claim that relied on it is automatically hidden and the page is flagged for review. Disproven information is never shown. A previously hidden claim stays hidden until a human re-approves it; we never silently re-publish.
7. Honesty enforced in code
The site won't build if a page contains an affiliate link, a fake citation, a prescriptive dose presented as a recommendation, a promissory “cures cancer” claim, marketing/SEO exhaust, a broken internal link, or agent data that drifts from our schema. These are automated checks that run on every build. The editorial standard is part of the code, not a hope.
8. Who reviews
We are not a clinical review board, and we don't claim to be. We are editors who read the sources and approve each page. Curated pages carry a visible “Reviewed” date; source-grounded pages show when they were last approved and when their sources were last checked. None of this replaces your medical team. It's a faithful, cited summary of the literature to bring to them.
Why this exists
OncoForge grew out of one family's search through the research after a rare, aggressive cancer diagnosis: read the story. It's free, ad-free, and sells nothing. See the live research feed and the cancer-type guides.