About Predictly

Between “I have a feeling that” and a reasoned estimate sits half a day of reading nobody has. To know whether a transfer will go through, whether a game will ship on the announced date or whether a measure will pass, you have to gather what came out this week, discard the copies of one wire story, work out who is speaking for themselves and who is relaying, and then make something of it.

That is the work Predictly does, and it shows it. You do not just get a percentage: you get what was read to arrive at it, and what cuts the other way.

You ask a question, in English or French, about something that is not settled yet. Predictly works out which event you mean, goes and reads what is being reported today, weighs it, and gives back:

• the most likely answer, in as few words as possible;
• its probability, rounded to the nearest 5 points;
• a confidence level — high, medium or low — saying how well documented the answer is;
• what argues for it and what argues against;
• the sources actually consulted, with their dates where known.

The calculation itself is on the methodology page, and how to read the percentage on understanding probabilities.

• It is not a betting site. It takes no stakes, offers no odds, and a probability is not an odd.
• It is not financial, investment or trading advice, and it must not be used to decide where to put money.
• It is not an oracle. Predictly estimates what is most likely to happen; it does not know what will happen.
• It is not a source of certainties. A forecast at 90 % is wrong about one time in ten, and that is normal.

An engine that answers everything answers some things badly. Three cases are turned away before any forecast is computed, and in each one Predictly says so rather than producing a number.

• An event that is already settled. If the resolution date has passed, or the question is asked in the past tense, Predictly says the event has already happened. It does not replay a known result as though it were a forecast, and it does not invent one either.
• An ambiguous subject. When the name in the question could mean two different things, Predictly names the readings and hands the question back. Researching both at once would give an answer decided by whichever had the busier news week — which is a forecast about neither.
• Too little evidence. Below two independent pieces of evidence from two distinct publishers, no forecast is made. Predictly says it does not have enough to go on.

Every source shown is a document that a search engine, a news feed or a reference database actually returned. The title, the link, the site and the date come from there, not from text a model wrote. That is a constraint in the code rather than an intention: a link no provider returned has no route onto the page.

That does not make the sources infallible. An article can be wrong; an announcement can be denied the next day. They are shown so you can check, not so you take them on trust.

That is the starting assumption, not a footnote. A probability is an estimate made from what was readable at the time of the research. An injury, a denial, a last-minute announcement, and the right answer changes. That is also why a forecast is re-dated and recomputed as it ages.

The only honest way to judge a forecaster is to look at many of its forecasts, not one. That idea has a page of its own: understanding probabilities.

The publisher, the host and the contact address are in the legal notice. What data is collected and how it is handled is described in the privacy policy, and the rules of use in the terms.