Ask a chatbot with web search whether something will happen, and you get a paragraph. It may be well written and it may cite sources. But it answers once, it rarely commits to a number, and nobody checks it afterwards. Prolepsia is a forecasting engine for teams. It is built to give a different kind of answer.
The answer is a number you can check
Prolepsia returns a probability for a yes-or-no question, a share for each option on a multiple-choice question, or a forecast range for a number or a date. The reasons and the sources it read come with it.
A number can be checked after the fact. If Prolepsia says 30% and the event happens, that is one data point. Across many questions, you can see whether the things Prolepsia called 30% happened about 30% of the time. A confident paragraph cannot be checked that way, because it never said how confident it was.
The evidence stays with the answer
Every forecast keeps the report that explains it and the list of sources Prolepsia read. On a public forecast you can read the report, open each source, and see every earlier forecast on the same question. If you disagree, you can point at the exact piece of evidence you would weigh differently.
The answer keeps up with the evidence
A chatbot's answer is fixed at the moment you asked. A Prolepsia forecast can refresh on a schedule you choose, or on a condition you set, such as a filing appearing or a number passing a threshold. The history on the forecast page shows how the probability moved and what changed each time.
The answer gets scored
Once an outcome is established, a forecast can be scored against what happened. Prolepsia uses the Brier score and log loss, two standard rules that reward a confident forecast that turned out right and charge more for a confident one that turned out wrong. Some questions cannot be scored, for example when no outcome is established or the question is annulled, so any record states what it covers and what it leaves out. How forecasts are scored explains the rules with worked examples, and the track record shows the scores of Prolepsia's public forecasts.
It uses your evidence as well as the public record
You can attach your own files and notes to a question, or connect a source over MCP, and label how far you trust each one. Prolepsia reads them alongside the public record: filings, registries, price series, official statistics and what the market already thinks.
When a chatbot is the better tool
For a quick explanation, a summary of a document or a first look at a topic, a chatbot is faster and costs less. A forecast takes longer, because Prolepsia researches the question and weighs the evidence before it gives a number. Use a forecast when you need a probability you can act on, defend and check later.
Try it
The public forecasts show finished examples. The example decision opens in the app without an account, and the docs show how to ask your own question.