Accuracy & Outcomes

Realised, not expected. Every figure here is what happened: contracts bought, stake paid, payout received, difference taken. That is arithmetic and needs no significance test, which is why nothing on this page is suppressed for having a small sample. It is also why no figure here may be read as an edge without checking how much data stands behind it. Polymarket's parlay series is the sharp case: it is % resolved across a product that launched on 2026-08-06, with an effective sample size of and a 95% interval of % to % of stake. Those bettors really did lose that money. Nobody should conclude from six days what parlays cost on Polymarket in general.

Whose P&L. Kalshi and Novig are the two venues publishing an aggressor flag, so their single-market numbers are the only true taker P&L here. DKeX, Polymarket US and ForecastEx publish one price per print against a symbol naming a leg, with no flag, so theirs is the P&L of whoever bought that leg — labelled named and never called taker. ProphetX is a third case, and it needed a third label. Its prints are two-sided exchange trades with a willing counterparty on each side and no aggressor flag anywhere in the feed, and the single price it publishes belongs to one nominated side — the second-named, home side of the fixture. That series is labelled priced_home: the away side’s P&L is the exact mirror of it, sign for sign, and calling either side the taker would invent a fact the venue does not record. The request-for-quote argument set out next, which makes Polymarket’s parlays genuinely taker P&L, does not carry over — nobody is lifting a venue’s quote here. Parlays are the exception: they are priced by request-for-quote, so the customer lifts a quote the venue makes and the taker is almost always the YES buyer. Polymarket's own feed carries that out exactly — Strike Price is YES on every parlay row, with no NO side listed — so its parlay series is genuinely taker P&L. The producer asserts that property on every run and fails rather than mislabel it.

Pricing basis. Every row carries its basis, and the table below renders that field directly.

What validates the method. Each exactly-priced venue is aggregated a second way and the two answers are compared on every build. DKeX’s folded series and a payout-minus-cost sum over the same calibration file agree to one cent on 21,485,019 contracts; ProphetX’s two published files, built by different aggregation paths, agree to $6 on $2.28M, or 0.0003%. Those are checks on the arithmetic, not confirmation of the resolution rules the two paths share — ProphetX’s rule is checked on its own terms instead, and its producer refuses to write anything at all if settlement polarity ever flips. Separately, Polymarket's parlay producer reconciles report and trade-tape volume on every run. The two sources use different date bases, so their price and resolution ratios are not presented as a like-for-like validation.

Kalshi's aggregate is held out of the chart. Kalshi publishes an all-markets series as well as its parlay and single-market halves; drawing all three would count the same contracts twice in one comparison. The aggregate is , and is quoted here rather than plotted.

Four venues were removed from this page on 2026-08-15, and this is why. A P&L answers “what did the bettors make”. Answering it requires knowing who traded — which side was the aggressor. Only Kalshi and Novig publish that flag per trade, and parlays are the one exception, because they are priced by request-for-quote: the customer lifts a quote the venue makes, so the taker is the buyer by construction. Without an aggressor flag a P&L is not what anyone made. It is a measurement of price bias, and its mirror side is its exact negation — if the priced side lost 1.19¢ per contract, the other side gained 1.19¢. That is a real and interesting quantity, but it belongs to the calibration curve above, where all four removed venues are drawn.

DKeX single markets — no aggressor flag (its tape is date, symbol, timestamp, price, quantity), so the named-side result is price bias rather than what the taker earned. Fees can reverse that result's sign, and customer cost now includes both the DKeX exchange charge and DraftKings Predictions' applicable introducing-broker commission; neither fact supplies the missing aggressor identity. DKeX parlays are different: they are RFQ-quoted, so their buyer is the taker, and that separate net series remains on the chart. ProphetX — no aggressor flag; its single-market number was the priced home side against its own settled outcome, which is price bias by another name. Polymarket US single markets and ForecastEx — neither publishes an aggressor field at all; ForecastEx matches a YES buyer against a NO buyer with no taker concept. Both were carried as named, which is the pipeline stating in its own schema that it does not know who traded.

Why the venues with big parlay books are still only partly here. Underdog, Novig, OG/Crypto.com and ProphetX all run substantial parlay volume, and request-for-quote pricing would settle who paid. For most of them the missing half is what happened. Underdog gives a closing price and a status that only ever reads “Finalized” — across 4,972 tickers not one ends at a settled 0 or 1, and of 282 two-sided game-win pairs, where exactly one side must win, none shows one winner and one loser. ProphetX carries no event date on any of its parlay contracts, so the maturity rule that makes its single markets safe cannot be applied. OG/Crypto.com publishes a daily bulletin with no traded price and no per-contract settlement. Novig is the exception: its single markets now settle to WIN/LOSS in the public GraphQL snapshot and are the taker bar on the chart above; only its parlays stay out, because 0 of its COMBO markets ever resolve.

Which venues are absent, and why. Underdog runs parlays — its UDXCOMBO is the large majority of its records — and publishes no settlement outcome, so no P&L is constructible for it and none is guessed. Novig's parlays are absent for the same reason (0 of its COMBO markets resolve), but its single markets settle and are the taker bar on the chart above. DKeX did list no parlay product until 2026-08-26, when it launched multi-leg combos; they are RFQ-quoted, so the buyer is the taker, and their P&L is on the chart above and broken out on DKeX · Parlays. ProphetX runs parlays too, and they are absent for a third reason: not one of the 80,543 distinct MULTI-EVENT- contracts in its bulletin carries a parseable event date, so the maturity test that makes its single markets safe cannot be applied to a parlay at all — 31,899,248 contracts left out, not estimated. Their terminal marks are no help either: of the 79,279 ProphetX parlays that settle to exactly 0 or 1, 94.92% mark to 1, which cannot be a multi-leg win rate, so those marks are not outcomes.

ProphetX: built, and deliberately not drawn as P&L. ProphetX has been described as a venue that records that a contract resolved but not which side won. That holds only for its parlays. On single markets the outcome is recoverable — a contract that has been delisted, whose last bulletin session falls on or after its event date, and whose terminal mark is exactly 0 or 1 — and a series built on that rule covers 918,445 prints and 191,575,583 contracts across 3,450 distinct fixtures over the 60 sessions from 2026-06-16 to 2026-08-14. It is not on the chart above, and that is a decision rather than a gap: ProphetX publishes no aggressor flag, so the series is the venue’s priced (home) side against its own settled outcome — a measurement of price bias whose mirror side is its exact negation, not what any trader made — and the pipeline removed every series of that kind from this fold on 2026-08-15. Measured on its own terms it finds no measurable edge: the priced side paid 44.15¢ on average and won 42.96% of the time, and that gap stands at t = 1.50 against a fixture-clustered standard error of 0.79¢ — it does not clear two — while zero of the 20 price bins clear two either, once each bin is measured against the price actually paid in it rather than against its midpoint. No per-contract figure anywhere on this site should be quoted as a ProphetX edge.

What ProphetX still cannot show. Even that unpublished series covers only 80.13% of ProphetX’s single-market contracts (191,575,583 of 239,095,073). The other 19.87% sits behind 194,030 prints that could not be joined to an outcome — 193,271 on contracts still unresolved at the end of the window, 657 on contracts never listed in the bulletin, 102 priced off-scale — and is excluded, not imputed. Its parlays are absent entirely: 31,899,248 contracts with no derivable outcome, for the reason set out above. So ProphetX has no bar on this page at all; its price-paid calibration is where its outcomes appear.

All curves are contract-weighted and use the strongest venue-wide settled group in each producer. Error bars cluster prints on the underlying event. Price-side conventions differ by venue and are documented in Methodology & coverage.