How the numbers are calculated

Every market shows a PredictX probability for each outcome: a display figure computed from the tokens committed to the market. Payouts are not computed from it — they come from the pool itself (see payouts).

Two-outcome markets (YES/NO questions, and head-to-heads such as one team against another) use Bayesian smoothing:

probability(A) = (tokens on A + k ÷ 2) ÷ (tokens on A + tokens on B + k)

k is a fixed smoothing weight, set per market type. Its effect is that a market with few tokens stays close to 50% instead of swinging to an extreme on the first prediction, and that the figure converges on the plain share of tokens as the pool grows. That is why, on a standard match market, 700 tokens on one side and 600 on the other reads 52%, not 54%.

Markets with three or more outcomes show each outcome’s share of the tokens committed to outcomes still in contention. An outcome that has been eliminated drops to 0% and its share is redistributed across the rest. Before any tokens are committed, the outcomes start level.

Some multi-candidate races are shown as a set of YES/NO markets, one per candidate, each smoothed on its own — on those pages the candidates’ figures need not add up to 100%. A race run as a single market with many outcomes follows the share-of-tokens rule above and does.

The PredictX probability is not a forecast by PredictX, an endorsement, or a guarantee. It is recalculated with every prediction, it can be wrong, and a market can settle against its favourite.

Some figures describe people rather than the pool: how many players have predicted on a market, how players split their tokens across the outcomes, whether the players’ favourite won, and how many players called the result. They appear in a market page’s description, FAQ, market information and result, and on the crowd accuracy report.

• Those figures are counted from player accounts only. Internal and test accounts are excluded. (Activity counters elsewhere on the site, such as the participant count on a market card, count every account.)
• They are shown only once enough players have predicted on a market. Below that, a “crowd” is one or two people, and a percentage would restate an individual’s choice.
• The players’ split and the PredictX probability are different measurements and are labelled separately. The probability is computed from the whole pool; the split is players’ own tokens. They usually point the same way, but they need not.
• The players’ “favourite” is the outcome that leads on both player tokens and number of players. Requiring both stops one large commitment from speaking for everyone else. When the two measures disagree, or two outcomes tie, there is no favourite, and no market page or report claims one.

PredictX markets are parimutuel pools. Every token committed to a market joins one pool. When the market settles, the pool is shared among the correct predictions at one base ratio per token:

base payout per token = (total pool − platform fee) ÷ tokens on the winning outcome

• The platform fee is 5% of a contested pool — one where some tokens were committed to a losing outcome — and nothing on an uncontested pool, where every correct prediction gets at least its tokens back.
• Two bonuses are added to the base payout on most markets: 10% for the first 10 predictions made within 5 minutes of a market opening, and 5% for a prediction held for at least 50% of the market’s lifetime. Some markets also carry a sponsor’s bonus pool, shared among correct predictions after settlement and labelled on the market.
• While a market is open, a prediction can be withdrawn for a full refund. Until it settles, a prediction can also be sold: it returns the tokens committed × that outcome’s current share of all the tokens in the market, less a 2% exit fee. That share is the raw pool share, not the smoothed PredictX probability, so the two can differ.
• The payout shown before you predict is an estimate based on the current PredictX probability. The payout itself is fixed only at settlement, from the final pool.
• Tokens are free, carry no monetary value, and cannot be bought, sold or exchanged for money.

• Every market settles against the real-world result of the question it asks. Where its resolution criteria have been written, they are shown on the market’s page.
• Markets settle in one of three ways: automatically from a live data feed that covers the event; by review against the official result; or, for some markets, automatically from published evidence when the result is unambiguous.
• For 24 hours after a result is published, players ranked in the top 100 can vote to dispute it; at 10 votes the result goes to review, and is either confirmed or reversed. A reversed result cancels the market, and every token committed to it is refunded.
• A market that cannot be settled — an event cancelled or abandoned without a result — is cancelled, and every token committed to it is refunded in full.
• A settled market’s page stays published with its result, its final PredictX probability and the date it settled. Once the dispute window has closed and any dispute raised within it has been decided, it does not change again, and it is the one to cite.

• An open market’s page is regenerated from the live market every minute or so, and a cached copy can occasionally be several minutes old. Signed-in players also see it update live. A market page states the time its probability was read; any figure quoted from an open market is accurate only as of that time.
• Times are shown in India Standard Time (IST, UTC+5:30) and labelled as such. The structured data on each page carries the same instants in UTC.

The crowd accuracy report asks one question of every settled public market with enough players: did the outcome players favoured win? It reports how often it did, broken down by how strongly players favoured it, by category and by competition, with the number of markets behind every rate.

• The players’ split is taken when the market settled, so it includes predictions made while the event was under way.
• Markets with no favourite are counted separately and not scored. Markets that settled with more than one winning outcome, and the per-candidate markets of a race shown as a set of YES/NO markets, are left out entirely — a race is one question, not one per candidate. Ball-by-ball and over-by-over micro-markets are left out too.
• A rate is shown only where enough markets stand behind it.

• Not a real-money platform. Nothing on PredictX can be deposited, withdrawn for cash or redeemed.
• Not a source of expert picks or tips. PredictX publishes what has been committed to each market, not what it thinks will happen.
• Not a forecast. A probability is a measurement of commitments at a moment in time.