Hi Product Hunt 👋 I built Predictly because answering "how likely is this, really?" takes half a day of reading that nobody has. You end up with a feeling instead of an estimate. You ask a question about something that hasn't happened yet. Predictly works out which event you mean, researches it across several sources, collapses duplicate reporting so one wire story isn't counted ten times, and answers with a probability — plus a confidence level computed from how solid the evidence is, what argues each way, and links to everything it read. Two things I care about: — Confidence and probability are kept separate. How likely the event is, and how well documented the answer is, are different questions. "15%, well established" is a coherent and useful sentence. — It refuses rather than guesses. Events already settled, subjects whose name means two things, and questions with too little evidence all get a plain refusal instead of a made-up number. What I'm not claiming: Predictly publishes no accuracy figure, because I have no verified record to publish yet. Calibration — checking whether things given 70% happen about 70% of the time — needs many resolved forecasts, and I'd rather build that in the open than quote a number I can't show the working for. What I'd love to learn from you: which questions you try, and where the answer felt thin. The refusals especially — if it declined something it should have answered, that's the most useful thing you can tell me.

Predictly

Research-backed forecasts for the questions ahead

Research-backed forecasts for the questions ahead