Prediction markets without the gamble.

Make a call. Learn something.

Your prediction stays editable until the question closes. After you save, the comparison shows other people’s average probabilities, excluding your own, alongside available AI forecasts and clearly labeled imported Polymarket snapshots. Illustrative fallback estimates never appear as AI in this comparison.

Resolved questions award up to 100 points using 100 × (1 − Brier score ÷ 2). Your last prediction before the deadline is scored, and results remain on record. Friend challenges use the same points, counting predictions submitted after joining. Daily goals and streaks follow UTC calendar days.

Creators publish answers with evidence after predictions close. Curated imported questions are checked daily against finalized source results; uncertain or void results are left unscored. Results appear in your in-app updates.

Try the future on for size.

Practice solo against an AI panel or invite friends to the same simulation. Everyone faces five outcomes randomly sampled from a frozen probability distribution. Your guess locks before results appear, and everyone is scored against the same outcomes.

These are simulated possibilities, not predictions of what actually happened. Practice scores are separate from real forecasting points. Beating the AI over five draws can be fun, but is not proof that you are a better forecaster.

01 · Purpose

Turn uncertainty into an inspectable estimate

Most difficult questions do not have a useful yes-or-no answer yet. They have evidence, disagreement, and a range of plausible outcomes. A prediction market compresses those competing views into probabilities that can be compared and revised.

A bounded model request asks eight forecasting perspectives to estimate the same outcomes independently. Their estimates are combined into one aggregate, while each perspective leaves a rationale that can be inspected alongside the result.

02 · Market mechanism

Panel estimates, normalized as probabilities

Each panel member supplies a complete probability distribution over the allowed outcomes. The app normalizes those distributions and combines them with equal weight, retaining the starting estimate as a prior. Confidence is displayed as context; it does not give one model persona extra voting power.

No money, tokens, or financial incentives are involved. The chart shows the aggregate after each panel update. If the live model provider is unavailable, the app clearly labels a bounded rule-based fallback instead of presenting it as live model analysis.

03 · Forecaster ensemble

Diversity is a feature, not decorative copy

Eight named archetypes represent common forecasting habits: base rates, trend signals, skeptical review, scenario construction, and contrarian pressure. Each makes a bounded update and leaves a short rationale in the activity log.

• Outside view

  Start with comparable historical cases before accepting a compelling narrative.

• Signal discipline

  Move on new evidence, but scale the update to its reliability and independence.

• Contrarian pressure

  Look for crowding, shared assumptions, and incentives that may bias consensus.

• Calibration

  Prefer honest uncertainty to false precision, especially when evidence is sparse.

04 · Decision decomposition

Separate the choice from the evidence

The Decide flow turns one broad question into editable actions, factors, and focused sub-questions. A second bounded analysis compares the reviewed frame, estimates the relevant uncertainties, and explains its recommendation.

This does not replace judgment. Its purpose is to expose the frame: which options were considered, what evidence would matter, and where uncertainty still lives.

05 · Data and limits

Shared by design, anonymous by default

Questions with people’s predictions or recorded results are retained to preserve the track record. Authors can delete unused forecasts, decision analyses, and their own comments while signed in or using their original guest cookie. Clearing an unlinked guest cookie removes access to its profile and author controls; signing in first keeps that history connected to your account. Invite-only practice rooms are accessible to anyone with their unguessable link, so share invitations thoughtfully.