How it works

A wrong verdict here is a public accusation against a named person, so we resolve predictions carefully. The site has one editor, who is accountable for every verdict. AI does the legwork, working under the editor's authority, and everything is published so anyone can check our work.

1. Someone clips a quote

A reader (or an editor) pastes a forecast — a tweet, an article, a clip — and it's archived immediately.

2. It becomes a falsifiable claim

The quote is turned into a specific, checkable claim with a deadline and a rule for what counts as right. The editor has to admit the claim before it counts — vague or unfalsifiable claims are rejected rather than scored.

3. Anyone can propose a verdict

Once the deadline passes, any signed-in reader can propose an outcome with evidence and a rationale.

4. AI drafts the case file, never the verdict

An AI dossier gathers evidence, quotes the relevant passages, and suggests an outcome — but it is only ever an input. It cannot finalise a verdict: every verdict goes through the editor and the 7-day window below.

5. The editor accepts it, with reasons

The editor accepts or rejects each proposed verdict and writes down why, citing the rule and the archived evidence. With one editor, the person who accepts a verdict may also have clipped the quote or drafted the proposal (often with AI's help), so no second person signs off. Two safeguards carry that weight instead: everything is public, and every verdict waits out the window below where anyone can contest it. The forecaster never decides their own verdict.

6. A 7-day provisional window

Every accepted verdict is public immediately but doesn't count toward anyone's score for 7 days — enough time for anyone to catch a mistake before it affects a scorecard.

7. Contests

During the provisional window, anyone can contest a verdict on a specific ground: the evidence was misread, the rule was misapplied, there's new evidence, and so on. A contest that holds up on its face blocks finalisation, and the verdict stays unscored until someone other than the editor who accepted it can review it. Until then it does not count.

8. No partial credit

Verdicts are "occurred," "did not occur," "void" (the claim itself was defective), or "unresolvable" (the evidence isn't there). There is no "partial" outcome — where a number is involved, we publish the actual figure instead of hedging.

Retrospective predictions

Some of our launch predictions were clipped after the outcome was already public — "hindsight clips." These are flagged and scored in their own cohort everywhere, so they never inflate a forecaster's real track record.

Quotes we get wrong

If a quote turns out not to be in its source, or was said by someone else, we take it down. It leaves every feed and score, and its page keeps a public note of when and why. If you're named on this site and think we misquoted you, email support@publicpredictions.markets.