How eight prediction-market venues count volume

Shares are not dollars

A prediction-market contract pays a fixed amount if its outcome happens, usually $1. Buying 100 contracts at 30 cents costs $30 but adds 100 to a volume figure that counts contracts. The two measures differ by the average price paid: a share bought at 10 cents adds ten times its cost to a share count, a share bought at 95 cents almost exactly its cost. Neither is wrong, but they answer different questions — how much risk changed hands, and how much money did.

Most venues do not say which they report. Their API documentation calls the field "volume" and stops there. pdata settled it where it could by adding up a market's individual trades from the venue's own public trade history and comparing the sum with the reported figure. The result held exactly for one market on each of Polymarket, Manifold, Myriad and Limitless; one market is a check of the definition, not a survey.

Polymarket

Polymarket's volume counts shares. For a 5-minute Ethereum up-or-down market, Polymarket reported a volume of 5,318.33; the 234 trades in its public trade history added up to exactly 5,318.33 shares, for $2,679.18 in cash. The trade history counts each fill once, from the side that took liquidity, and its sum matched the volume figure to the share, so each trade is counted once rather than once per side.

pdata reads the lifetime figure (volumeNum) and the 24-hour figure (volume24hr) for every market and event, and passes them through unchanged. An event's figure is Polymarket's own, and in the data pdata has captured it equals the sum of its markets. Polymarket's one-week and one-month fields are a trap: pdata found in August 2026 that they reset a few days after an event closes, leaving 3.1% of the true monthly figure across a 30-day group of closed events, so any monthly total built from them undercounts badly. pdata does not use them.

Kalshi

Kalshi documents its volume as a count of contracts, and its contracts are worth $1.00 each at settlement. pdata reads volume_fp and volume_24h_fp per market. Kalshi's events carry no volume of their own, so an event's figure on pdata is the sum of its markets. A check against one market's public trades came within 2.3% — 126.84 contracts in the trade history against a reported 129.84 — and the gap is unexplained; older trades may sit in a separate history endpoint.

Two things are missing from Kalshi totals on pdata, by design. Combination bets do not appear: Kalshi's events endpoint does not return them. And Kalshi's intraday price markets — 15-minute contracts on crypto, commodities and currencies, and hourly crypto, stock-index and currency contracts — are dropped when pdata ingests Kalshi, because a 24-hour figure for markets that settle every quarter hour cannot be measured accurately without every trade.

Manifold

Manifold runs on play money, mana (M), which has no cash value. Its volume is the mana spent on purchases plus the mana received from sales; automatic payouts when a market resolves are not counted. On one market checked, the reported 400.85 equalled 332.46 spent on buys plus 68.40 received on sells, with 31.03 of redemptions left out.

Manifold's multiple-choice questions report a lifetime volume for the question that can be several times the sum of its answers' volumes — 2.8 to 14.2 times in three examples pdata captured — and its documentation does not say why. pdata keeps the question's figure, computes each answer's 24-hour volume from its own records, shows mana as M rather than $, and leaves Manifold out of the homepage's volume chart and its API grand total.

Myriad

Myriad reports two volumes, and pdata uses the one called volumeNotional. Despite the name, it counts shares: on one market it was 753.35, exactly the number of shares bought and sold, while Myriad's other field, volume, was 388.66 — exactly the cash. pdata keeps only markets settled in a stablecoin on the BNB Chain and Abstract networks, and leaves out Myriad's points markets.

Limitless

Limitless reports a lifetime volume in its position tokens — shares — with six decimal places: one market's 21.75 matched its two trades' 21.75 shares exactly, for about $6.00 in cash. Limitless publishes no 24-hour figure, so pdata computes one from its own records of the lifetime figure: the sum of every increase over the past 24 hours. That is an estimate. Trading between two of pdata's readings is counted when the next one arrives, so the edges of the window are blurred by up to one polling cycle, and a market less than a day old shows its whole lifetime volume as its 24-hour volume.

Predict

Predict's lifetime volume comes from a per-market statistics endpoint, in a field named volumeTotalUsd, and its markets are settled in USDT on BNB Chain. Whether the figure is cash or shares could not be checked: the endpoint requires an API key and the documentation gives the field no description. Predict's own 24-hour field is empty for about 95% of markets, so pdata computes 24-hour volume the same way as for Limitless. pdata can refresh statistics for 3,000 Predict markets per cycle, busiest first, so a quiet market's lifetime figure can lag.

Opinion

Opinion reports volume in USDT, the stablecoin its BNB Chain markets settle in. Whether that is cash or shares is not documented and was not established. One thing was: for a question with several answers, Opinion's question-level volume is exactly twice the sum of its answers — 59.958 against 29.979 and 221.78 against 110.89 in the two examples captured — which looks like each trade counted on both sides. pdata therefore halves Opinion's question-level 24-hour figure and builds the lifetime figure from the answers.

Gemini

Gemini's documentation describes its event volume as "total trading volume in USD". The data suggests contracts: on every event pdata captured whose contracts trade in whole units, the volume was a whole number — 8 of 8 — which dollar amounts traded at cent prices would rarely produce. Each Gemini contract is worth $1, so the two readings may be the same thing described two ways. Gemini reports volume only for whole events, so pdata shows no per-contract figure.

How pdata reports them

pdata converts none of these figures. Each venue's number is shown in that venue's own unit, which is what makes it checkable against the venue's site. The homepage puts a $ sign on every venue's figure except Manifold's, as Polymarket and Kalshi do on their own sites for contracts paying $1; read it as "contracts" rather than "dollars paid". Shares of a venue's volume by category are comparable across venues; sums across venues are only an order of magnitude, which is how the API describes its grand total.

Two of pdata's own numbers were wrong until this check and are fixed: the methodology page said only Limitless had its 24-hour volume computed by pdata, when Predict's is too, and the Manifold hub page showed mana with a $ sign. The coverage report describes a third, larger correction, to Kalshi's 24-hour total.