These standards apply to every claim, verdict and page on SiPoly.
Verbatim quotes from primary sources
- Every quote is copied word for word from a primary source, such as the person’s own post, a transcript, a filing or an official announcement, or from reputable reporting of it. We never paraphrase inside quotation marks.
- We save a copy of each source when we collect a quote, and we link an archived copy (for example on the Internet Archive) wherever one exists, so you can check the original even if it changes or disappears.
- Dates are only as precise as the source. If we only know the year, we show the year.
- We do not rely on anonymous sources to establish what someone said. If a verdict draws on reporting that cites unnamed sources, the grader’s note says so.
Criteria first, grading later
- For every claim we write down how it could be settled and by when, before it is graded. These resolution criteria are our interpretation, and they are shown on the claim page.
- When the date arrives, we grade the claim against those criteria and the public evidence. We grade the claim, not the person.
- Verdicts use fixed labels: correct, partly correct, wrong, too vague to judge or withdrawn (for promises: kept, partly kept or broken). Vague or hedged claims count for less in scores. See our methodology.
A second review before anything is published
Nothing goes live on research alone. Each claim and verdict is drafted, checked automatically, and then reviewed by a human editor, who confirms the quote against the source, the date, the fairness of the criteria and whether the evidence supports the verdict. Only then is it approved and published. Anything still waiting for that review is marked “pending editorial review” and kept out of search engines.
Corrections and right of reply
- If we get something wrong, we fix it and note the change and its date on the claim page.
- Anyone we quote, or someone speaking for them, has a right of reply. We may summarize or quote replies on the claim page.
- Anyone can send a correction or new evidence through our corrections form. Our goal is to review every request within seven days; complex cases can take longer.
- A verdict can change when new evidence appears. Each claim page shows when it was last reviewed.
Neutrality and balance
- We actively seek out skeptics as well as optimists, critics as well as labs, and apply the same rules to all of them.
- We write grader’s notes in plain, neutral language.
- We only include public statements by public figures, companies and public bodies, never private individuals.
Independence, funding and conflicts of interest
SiPoly is an independent project published by Bhanu P. It accepts no payment from the people or organizations it tracks, and advertising, if any, never influences which claims we track or how we grade them. No one can pay to add, remove or change a claim or verdict.
We know of no conflicts of interest. If one arises, such as a financial or personal tie to a person or organization we track, we will disclose it on this page and on the claims it affects.
Use of AI
Our research and drafting are AI-assisted. AI tools help us find candidate claims, gather sources and draft criteria and grader’s notes. Every quote is machine-checked against the saved copy of its source, and every verdict is approved by a human editor before it is published.
How we choose claims
- The statement is about AI and was made in public: in a post, interview, talk, paper, filing or official announcement.
- It comes from someone whose words carry weight in the AI debate, such as lab leaders, researchers, investors, critics, companies and governments.
- It can be checked, or we can explain clearly why it cannot.
- We can find the exact words in a primary or reputable source. If we cannot verify a quote, we leave the claim out.
- We aim for balance across viewpoints and topics. Suggestions are welcome through our contact form.