How Cierto forecasts work
Cierto uses Jev AI to estimate the chance of events from public posts, news articles and official reports. These are model estimates. They do not confirm that an event will happen.
Each question has answer conditions
A question defines the event and what counts as an outcome. Some questions cover a UTC calendar day. Others cover a rolling period or a scheduled event. Read the conditions on the question page before comparing forecasts.
From public sources to a forecast
The system collects reports from the sources assigned to each question. It checks report dates, removes exact duplicate text and uses AI to check relevance. It sends selected excerpts and the answer conditions to Jev. Source limits and failed collection can leave gaps.
The sources behind an answer list all reports used for that forecast, highest relevance first. Reports with the same score appear newest first. Each report shows its relevance score as a percentage. Reports with scores of 80% or more are highlighted. Older forecasts may include reports that were not scored; these appear last. For new forecasts, relevance measures how useful a report is for the question at the forecast time, considering its age and the target period. Earlier announcements can remain relevant when they describe an upcoming event. Older forecasts can retain scores from the previous method. Relevance does not measure the chance of the event or prove that the report is true.
Browse collected reports to inspect the source material. A collected report is not necessarily evidence used in a forecast.
What the answer and probability mean
For a yes-or-no question, the page shows MAYBE when the estimated probability of YES is between 40% and 60%, including both limits. Outside that range, it shows the more likely answer. Questions with several options show the selected option and its probability.
A 70% estimate still leaves a 30% chance of another outcome. A model estimate is not a measured error rate. Cierto does not claim that its forecasts have proven accuracy across these topics.
Updates, expiry and history
Collection and forecast updates run separately. Each scheduled forecast reassesses relevance and report age, even when the collected text has not changed. The answer date records when Jev produced that answer. Collecting reports alone does not create a new answer date.
Expired forecasts stop appearing as current answers. The history charts show estimates over the last seven days. Some history records are retrospective estimates generated later; they are not predictions recorded before those events.
Limits to keep in mind
Public sources can be incomplete, delayed, wrong or misleading. The system does not see every report. AI can miss context or assign poor probabilities, especially for rare events. Check the linked reports and current official guidance before making an important decision.