Predictions should leave receipts.

Why this exists

Predictions are cheap because they’re rarely kept. Pundits move on; confident calls that turned out wrong quietly disappear. A forecast becomes more useful when there is a public record of what was believed and how that belief changed.

That’s the part atproto makes possible. Forecasts made through Matakite are published to your public, content-addressed repository, whose commits are cryptographically verifiable. Each revision made through the app is a separate record, and Matakite keeps the revision and retraction events it observes in your track record. Your PDS remains the authoritative source; Matakite is the index that turns those records into a usable forecasting history.

How forecasting works

A forecast is a specific, testable prediction you’re later held to. Matakite deals in resolvable questions — ones with a clear rule for deciding the outcome — and you answer them with probabilities rather than flat calls. Real predictions are rarely certain, so you say how likely an outcome is, not just whether it happens: “70%” carries more than “probably.”

Questions come in a few shapes, and each asks for that likelihood a little differently:

• Binary — a yes/no outcome; you set how likely “yes” is, from 0 to 100%.
• Multiple choice — a set of mutually exclusive options; you split 100% across them.
• Numeric — a number in a range (say, a vote share); you place a distribution over the range rather than guessing one value.
• Date — when something will happen; you spread your belief across a window of time.

Across many questions, what matters is calibration: if the things you call 70% likely happen about 70% of the time, you’re well-calibrated. That rewards being both accurate and decisive — confident and right beats confident and wrong, while hedging everything to “maybe” is safe but says nothing. You can revise a forecast as things change, and the whole trail is scored over the life of the question, so calling it early counts for more than piling on once the answer is obvious. Skill shows up as a track record over many resolved questions, not one lucky call.

Your reputation is yours

Matakite computes your calibration score from public records associated with your DID. Another implementation can read the same records and recompute the score under the public scoring specification; it does not depend on a private Matakite account or an opaque rating. Matakite is a convenient lens on the data, not the sole authority on your track record.

Why no money

Matakite isn’t a prediction market — you don’t trade, bet, or stake anything. Research on forecasting tournaments has found that prediction polls can be competitive with, and in some studied settings outperform, prediction markets when forecasts are aggregated using performance weighting, recalibration, and temporal decay. Leaving money out lets Matakite focus on forecast quality and public track records. Read the study.

Getting started

See the Docs for how questions, forecasting, and scoring actually work. Or sign in with an AT Protocol account, including a Bluesky account, and record your first forecast.