In the same month, takers paid $234 for sub-cent contracts on one Angels–Giants baseball game. Those contracts paid $206,238.

Both examples come from July purchases that settled within 30 days, the window we use for returns throughout this study. They show how little a contract count alone tells a reader about the money at stake or the eventual payout. We looked across the third quarter of 2026 to measure how much buyers spent on contracts below one cent and what those purchases returned if held to settlement.

The purchases that settled within 30 days lost money in July and September, gained in August, and lost 22% over the quarter. Giving every event equal weight produces a large gain instead, though most of it disappears when we leave out events where buyers spent less than a dollar. Under either measure, the result depends heavily on a few events.

We read the Polygon logs of Polymarket’s international exchange for every block from July 1 to September 30, 2026 (UTC), covering 155,226,449 taker fills. We kept purchases below one cent at the exact price of each matched order and used the recorded settlements to calculate their payouts. Throughout the piece, “paid” includes the taker fee. The returns assume buyers held until settlement; we don’t follow wallets, so we can’t tell whether they sold earlier. (Source and method.)

Explore and download the study data.

Lots of contracts, little money

Over the quarter, takers bought 1.05 billion contracts below one cent. That is 7.7% of the 13.7 billion contracts that takers bought at any price. They paid $2.70 million for them, or 0.041% of the $6.63 billion that takers paid in all. A sub-cent contract cost about a quarter of a cent on average, fee included. The average contract at any price cost 48.5 cents.

The gap held in every month:

• July: 10.1% of contracts bought, 0.055% of money paid

• August: 5.2% of contracts bought, 0.026% of money paid

• September: 6.0% of contracts bought, 0.034% of money paid

Figure 1. Below one cent, takers bought 5.2% to 10.1% of all contracts each month, but paid less than 0.06% of the money. Paid includes the taker fee. Taker buys on Polymarket’s international exchange, July to September 2026. Source: Polygon / Polymarket, Analysis: Splitrule. Values.

Across 46,221 events with sub-cent buys that settled within 30 days, half had less than $1.77 spent on those buys over the whole quarter. Most of the money was concentrated elsewhere: the 2.9% of events with $100 or more in settled sub-cent purchases accounted for 86% of the total.

So a large contract count at these prices usually stands for very little money. That matters for anything built on volume: a heat map of activity, a “most traded” list, or a claim that traders are piling into long shots.

What the money did

To calculate returns, we use purchases whose market settled on chain within 30 days of the buy. Here is how the quarter’s $2.70 million of sub-cent purchases breaks down:

81.4% settled within 30 days. Only this part enters the returns below.

17.9% ($484,639) did not settle within 30 days. By November 3, when our settlement scan ends, $105,122 of it had settled. For the other $379,517, our scan found no settlement. It includes long-dated markets such as the 2026–27 Champions League winner, which resolves in 2027.

0.65% ($17,641) was recorded after its market’s settlement in our scan. Those buys stay out by rule.

The venue’s token list lets us match every sub-cent buy to a market condition. Settlement coverage is less complete: our scan found no on-chain settlement for some markets that the venue shows as resolved. (The 30-day rule.)

By month: the share of sub-cent money settled within 30 days, the return held to payout, and its central 95% resampling range.

• July: 88.0% settled, return −41.3% (range −81.6% to +47.0%)

• August: 69.2% settled, return +76.7% (range −17.0% to +202.1%)

• September: 74.3% settled, return −38.3% (range −71.8% to +32.2%)

• Quarter: 81.4% settled, return −22.3% (range −59.8% to +45.5%)

Figure 2. Held to payout, sub-cent buys settled within 30 days lost money in July and September and gained in August. Every resampling range includes zero. The hollow marks are scenarios for the buys that did not settle within 30 days: worth nothing, or paid at the settled payout per contract. Source: Polygon / Polymarket, Analysis: Splitrule. Values.

Resampling the quarter’s events produces ranges that cross zero in every case, reflecting how much the results depend on the events in the sample. The calculation treats each event as a separate draw, although markets on the same asset and day can be related. Another resampling run moves the endpoints by several percentage points. These ranges describe sensitivity to the sample; they don’t account for the study’s other limits or predict future returns. (Resampling ranges.)

We also checked how much the 30-day window affects the result. If the purchases that did not settle within 30 days are worth nothing, the quarter’s return falls to −36.3%. If they pay the same amount per contract as the settled purchases, it is −25.2%. For the quarter to break even, they must pay 3.3 times that amount per contract. August stays positive even if all of them are worth nothing.

Part of the answer is already known: the $105,122 that settled after 30 days, but by November 3, returned −51.6%. Counted in, it moves the quarter’s return to −23.6%. We keep the 30-day result as the main one, because we fixed that rule before we saw any return. (Purchases not settled in 30 days.)

Two ways to average

There is a second way to average the same purchases: give each event the same weight, whatever was spent in it. On that basis, sub-cent buying gained in every month, though July’s resampling range includes a loss.

• July: −41.3% per dollar, +30.6% per event

• August: +76.7% per dollar, +66.4% per event

• September: −38.3% per dollar, +190.3% per event

• Quarter: −22.3% per dollar, +92.8% per event

How can the same purchases lose 22% and gain 93%? Most events paid nothing: 98.9% of them over the quarter. Each of those counts as a loss of 100%. The 530 events that did pay returned, on average, 168 times what buyers spent in them. 1.15% of events times 168 gives 1.93, or a gain of 93% per event. The per-dollar figure follows the money instead. 86% of the money sat in the 1,329 events with $100 or more of sub-cent buys, and as a group they lost 31.3%. (Returns.)

Figure 3. In July, September, and the quarter, the per-dollar and per-event returns have different signs. The per-event gain falls sharply when events with less than $1 of settled sub-cent buys are left out. We chose that floor after seeing the results, to test sensitivity. It is not a trading rule. Source: Polygon / Polymarket, Analysis: Splitrule. Values.

About a third of the events had less than $1 of settled sub-cent buys, together accounting for just 0.18% of the settled money. Yet they supplied 61% of the sum used to calculate the per-event average: each event’s payout divided by its spend. That share refers to gross payout multiples, before subtracting the purchase cost to calculate the net return. Without them, July falls to −2.1% and August to −4.7%. September falls from +190.3% to +67.7%. The per-dollar return moves by less than half a point.

Figure 4. Events with under $1 of settled sub-cent buys were about a third of the events and almost none of the money, but most of the sum of payout multiples behind the per-event average. Events with $100 or more held most of the money. Source: Polygon / Polymarket, Analysis: Splitrule. Values.

Raise the floor further and the per-event figure swings. For the quarter, it is +49.4% when we keep only events with $10 or more of settled sub-cent buys, and +201.8% at $100 or more, on fewer events each time. The tested floors give very different averages. The typical event, the median, lost everything at every floor and in every month.

We chose these floors after seeing the results to understand the average’s sensitivity. Testing a trading rule such as “buy only where others spent more than $100” would require knowing what a buyer could observe and purchase at the time. (Spend floors.)

A few events decide the answer

Half of the quarter’s sub-cent payout came from 7 of the 46,221 settled events. In July and in August, three events held more than half of the month’s payout.

Figure 5. Half of each month’s sub-cent payout came from three to eight events, out of about 15,000 settled events a month. Source: Polygon / Polymarket, Analysis: Splitrule. Values.

The monthly results depend on those extremes. Removing each month’s three biggest winners leaves every month with a loss; removing its three biggest losers leaves every month with a gain. Those one-sided removals deliberately test the influence of the extremes. Remove the three biggest winners and the three biggest losers together, and every month keeps its sign: July −57.2%, August +7.0%, September −23.3%, and −18.0% for the quarter. (Concentration.)

The winners and losers look different. The largest payouts mostly came from small buys: the Angels–Giants game, the World Cup Golden Ball winner, a market on Bitcoin’s price on August 19, and two women’s tennis matches. The largest losses came from long shots in big events: the World Cup winner (−$412,848), the September Fed decision (−$206,160), and the July Fed decision (−$121,626). Part of that contrast comes from the loss ranking itself: an event cannot lose more than was spent in it, so the largest losses come from events with a lot of money.

The broad monthly results also hide large differences between groups of purchases. In August, the quarter’s best month, markets in the sports fee category lost 30.7%, while markets in the crypto fee category returned +360% and supplied 69% of the month’s settled payout. In July and September, both losing months, the cheapest buys, below 0.2 cents, gained 36.9% and 83.5%. That cheapest slice gained in all three months, but it rests on a few events as well. In July, three events supplied 93% of the slice’s payout. Without each month’s three biggest winners, the slice lost money in every month: −90.4% in July, −59.3% in August, and −4.9% in September. That dependence on a few winners limits what we can infer from the cheapest purchases’ three positive months.

How the markets settled matters as well. In September, 34.6% of the settled payout came from markets whose proposed result was disputed at least once before it settled, according to the venue’s record. Markets that settled 50-50 supplied 14.3%, almost all of it from markets that were also disputed. A 50-50 settlement pays half a dollar per contract on either side, giving even the long-shot side a payout. Without both groups, September’s return is −59.6% instead of −38.3%, and the cheapest slice lost 27.8% instead of gaining 83.5%. In July and August, the two groups held 2.6% and 1.4% of the settled payout. For the quarter, the return moves from −22.3% to −22.8%. The dispute record establishes that someone challenged a proposed result. Whether the final result was wrong, or what a buyer knew, remains outside this evidence. (Settlement history.)

Sidebar: did late buyers drive the result?

Some markets turn on a fixed moment, such as a crypto price at a set minute. For those markets, we could find buys recorded on chain after that moment and before the settlement was recorded. These purchases accounted for 1.3% of the settled sub-cent money. Their block times record when trades reached the chain, after an unknown delay from matching off chain. They lost 98.3% in July and 99.4% in August. In September, one market changed the picture: in the Solana Up or Down market for 6 AM ET on September 16, $22.80 of such buys paid $21,231. The venue’s record shows that a proposed result for this market was disputed before it settled. Removing every such buy moves the quarter’s return by less than one point.

Figure 6. We could test buys recorded on chain after a fixed cutoff for only 9.6% of the settled money. Those buys were 1.3% of it. The remaining 90.4% has no established cutoff. Block times record arrival on chain; matching times and buyer knowledge remain unknown. Source: Polygon / Polymarket, Analysis: Splitrule. Values.

Among the markets without an established cutoff were the Angels–Giants game and a women’s tennis match, two of July’s three biggest payouts. In both, every settled sub-cent buy reached the chain less than an hour before the recorded settlement. Without the matching times or a fixed cutoff, this timing check cannot classify those trades. (Timing check.)

What to ask of a cheap-contract claim

Ask for dollars as well as contracts. Below one cent, 7.7% of the contracts were 0.041% of the money.

Ask which average. Per dollar, the quarter lost 22%. Per event, it gained 93%. The median event lost everything.

Ask how many events carry it. Without the three biggest winners, every month loses money. Without the three biggest losers, every month gains.

Ask how the markets settled. In September, 34.6% of the settled payout came from markets with a disputed proposal, and 14.3% from 50-50 settlements.

Ask what is left out. About 18% of the sub-cent money had not settled within 30 days, and for 14%, our scan found no settlement by November 3.

Ask whether the fills were there for you. A held-to-payout return describes fills that other buyers already took. It does not show what a new order can buy at that price.

The difference between contract counts and dollars runs down to individual fills: 94.5% of the taker fills with sub-cent buys spent less than $1 on them, together accounting for only 4.1% of the sub-cent money. Before treating a large count as a signal, a reader needs to know how much was spent, how the markets settled, and which events produced the payout. This quarter shows how much those details can change the answer.

The distribution map is broader than those two rivals. Crypto.com’s exchange, CDNA, supplies contracts to several sports brands and to its own OG app, and ForecastEx reaches customers through Interactive Brokers (IBKR), Lumina, and Robinhood. Some apps offer several providers while owning an exchange themselves. PrizePicks, which lists Kalshi markets today, has also announced a Polymarket partnership. A distribution deal can therefore expand an exchange’s reach without securing all of an app’s trading activity. The published routes show where that competition takes place; they do not tell us how much volume each relationship delivers. [1], [2]

Explore the companion map for the market access, ownership, clearing, and liquidity relationships behind these apps.

ROLR, the newest app on the map, launched on October 21. Behind High Roller’s customer app sits a chain of brokers and market infrastructure. ROLR US introduced customers to OG Markets US, doing business as Crypto.com FCM, which carried their accounts and funds. The contracts came from Crypto.com | Derivatives North America, or CDNA. [3]

High Roller also secured source-code rights and development support from DeepEther Labs, the business behind mrkts. That arrangement let it build the customer experience using licensed software while relying on other firms to carry accounts and operate the exchange. [4]

ROLR gave CDNA another channel for trading activity and gave High Roller a product for its audience. Their June agreement made CDNA the exclusive contract provider on High Roller’s US platform for the agreement’s first 24 months and anticipated revenue sharing, although it gave no percentage. [5]

A launch like this creates competition for the customer relationship around an existing trading pool. The app can attract users and earn revenue from their activity while the exchange continues to provide the place where orders meet. Following those relationships helps explain how so many prediction-market brands can emerge around a smaller set of venues.

The customer relationship has its own economics

Lumina discloses how this arrangement works in unusually clear terms. It uses IBKR exclusively to carry accounts and ForecastEx exclusively for trading. ForecastEx belongs to IBKR Group, and Lumina and IBKR Group share common individual control through Thomas Peterffy. [6]

Consider 100 ForecastEx contracts bought at 40 cents each. Lumina charges a one-cent exchange fee per contract, plus a commission of 0.35% of trade value. IBKR’s published direct-client schedule, which can vary by account, lists the same ForecastEx exchange fee with no commission. [7], [8]

Of the dollar in exchange fees charged to the customer in this example, Lumina receives 25 cents. IBKR also retains an undisclosed portion of Lumina’s commission, so the published charges reveal more about the customer’s cost than either firm’s eventual profit. [6]

The extra 14 cents pays for Lumina’s app and service on top of a venue that IBKR clients reach directly.

The exit can matter as much as the entry. Webull charges one cent in commission and one cent in exchange fees per contract on each trade. In a hypothetical round trip, 100 contracts bought at $0.40 cost $42 with fees. Selling all 100 at $0.44 returns $42 after fees: a $4 gross gain becomes zero after $4 in transaction charges. Holding through settlement has no additional settlement fee under the same schedule. [9]

Liquidity is a separate relationship

In institutional trading, a venue and a liquidity provider can each play a part in the same transaction. Cantor’s August announcement describes its introducing-broker role for Kalshi block trades, where clients can negotiate a single price away from the central order book. Susquehanna Predictions provides pricing and liquidity for that service. [10]

BitGo’s March offering gives customers a bilateral, over-the-counter event-linked derivative with BitGo as their counterparty and Susquehanna Crypto providing liquidity. Customers can post digital assets as collateral, though the disclosure leaves BitGo’s hedge venue unidentified. [11]

Susquehanna also appears in Robinhood’s Rothera joint venture, which is a different relationship from providing liquidity for a customer trade. A map that labels all of these connections as partnerships would lose the distinction between owning infrastructure and supplying prices. [12]

Polymarket’s builder documentation describes app orders matching through its central limit order book. The app’s attributed activity can therefore be the same trading activity recorded by the exchange, which matters when adding up volume across brands. [13]

Betr uses the separate Polymarket US book and usually displays its bid-ask midpoint as the probability, switching to the last trade when the gap exceeds ten cents. A customer comparing its screen with another app needs the price available for their order to judge the cost of trading. [14]

Shared software can serve different pools

Matchbook’s B2B customers can use its exchange technology with either shared liquidity or a ring-fenced market structure. That choice allows products built on the same software to offer different pools of counterparties. [15]

Its World Cup campaign provides a named example: PredictStreet used Matchbook technology and shared an account and wallet with its exchange. The June–July campaign establishes that deployment, while leaving its current operating status and precise degree of order-book sharing unconfirmed. [16]

Plaee combines software with access to CDNA’s contracts and liquidity. BitMart US announced a Plaee-powered product embedded in its platform, but its customer announcement leaves the exchange and carrying entity unnamed. The software connection alone doesn’t fill in those missing parts of the trading arrangement. [17], [18]

Plaee’s announced ProphetX integration, still described as beta, adds another potential venue to its offering. Each operator’s deployment would need to show whether its customers actually use that connection. [19]

Ownership expands the choices

Brands can own infrastructure while continuing to distribute outside contracts. Several published arrangements show both:

Robinhood. Its Rothera affiliation coexists with Kalshi, ForecastEx, and CDNA access. It also said it will hold equity stakes in Crypto.com and OG.com after OG’s spin-off. [20], [21]

DraftKings. It owns DKeX; its terms also name CME and CDNA. [22], [23]

Underdog. It owns UDX while retaining Kalshi and Nadex (CDNA) access. [24], [25], [26]

Fanatics. Its Water Street Labs and CX Clearinghouse acquisition closed August 5. Published retail terms still name CDNA; a migration is unverified. [27], [28]

Owning an exchange gives a distributor more control over the economics of a trade, while outside venues can supply contracts or liquidity its customers still want. These businesses have kept both options available, though their disclosures don’t show how trading activity divides among them.

The firms responsible for customer accounts and clearing add another layer to these arrangements. An account provider holds funds and positions, while a clearinghouse, or DCO, manages settlement and the obligations between trading parties. Public identifies both its own entity and Apex Clearing as futures commission merchants, or FCMs; DraftKings names Wedbush as a clearing FCM that acts through the applicable clearinghouse. [29], [23]

Gemini operates the Titan exchange. Its affiliated Olympus clearinghouse began operating on August 4, according to Gemini, bringing those functions into the same corporate group. Its nonbinding letter of intent with Apex proposes another source of customers through Apex’s FCM. Novig uses a different structure, with Ludlow operating the exchange and external Bitnomial Clearinghouse providing clearing membership. [30], [31], [32], [33]

Two applications now pending at the CFTC take different routes to clearing. Nasdaq MRX Futures applied on October 26 to become a designated contract market, and its proposed rulebook names the Options Clearing Corporation as its clearing corporation. OCC is an established clearinghouse that Nasdaq shares with other exchange groups, so the new market would plug into shared infrastructure in which Nasdaq already has a stake. The proposed contracts are binaries on year-over-year headline and core CPI that would pay $100 or nothing. [34], [35], [36]

Forum Markets Group has applied for both an exchange, ForumEX, and a clearinghouse, Forum Clearing, and owns each outright. Forum describes itself as an attention market, and its proposed contracts work differently from a binary that pays once: they would settle against an index at the end of every trading day and, by default, roll into the next day’s contract. [37], [38], [39], [40], [41]

Coinbase completed its acquisition of The Clearing Company in January, and Coinbase Clearing LLC registered as a DCO on September 28. Coinbase’s consumer help still names Kalshi as the market operator, so Kalshi remains the retail route customers can verify today. [42], [43], [44]

Data, wallets, and resolution add other relationships

ICE distributes Polymarket’s market information without handling customer trades. It launched normalized data feeds in February through its existing data services, whose terms explicitly exclude trading on or linking through to Polymarket. Alongside ICE’s investment in the company, those feeds create a commercial relationship around the information the market produces. [45], [46], [47]

Phantom’s disclosure makes the date a position opened important to tracing its route. Positions opened before June 1 used Kalshi through DFlow; later positions use World, which routes to Solana liquidity providers and redeems automatically. The disclosure leaves those providers’ hedge venues unidentified. [48]

Resolution adds another relationship to trace. Polymarket’s documentation distinguishes UMA resolution from Chainlink TWAP resolution for up/down markets, so even contracts on the same venue can use different mechanisms to determine a payout. Sharing one of those services tells us how an outcome is determined, while the contract’s own rules still define what the customer holds. [49]

For a customer choosing between apps, the practical comparison starts with the contract and the rights it carries, then the executable price, available size, and fees on entry and exit. The same venue can cost different amounts through Lumina and IBKR, and Matchbook’s choice of pools means shared software doesn’t establish shared liquidity.

For Kalshi, Polymarket, and the exchanges competing with them, much of the contest now runs through these relationships. A new app can become an exclusive channel, as ROLR is for CDNA under its current agreement, offer contracts from several providers, or build an exchange of its own while keeping outside access. ROLR’s app owns the customer relationship and shares in revenue, while other firms supply its software, accounts, and contracts. The companion map follows those connections so readers can see which parts of a trade each business controls.

References

1. PrizePicks: announced partnership with Polymarket

2. PrizePicks: prediction-market exchange and regulatory documents

3. High Roller: launch of the ROLR prediction-market app

4. High Roller: prediction-market software agreement with DeepEther Labs

5. High Roller SEC exhibit: CDNA exclusivity and revenue-sharing agreement

6. Lumina: conflicts of interest, IBKR relationship and revenue sharing

7. Lumina: exchange fees and trading commissions

8. Interactive Brokers: event-contract commissions and exchange fees

9. Webull: event-contract trading and settlement fees

10. Cantor: institutional Kalshi block trading with Susquehanna liquidity

11. BitGo: institutional prediction-market offering and liquidity arrangements

12. Robinhood: World Cup contracts and the Rothera joint venture

13. Polymarket: builder program and order-book integration

14. Betr: how prediction-market prices are calculated

15. Matchbook: exchange technology and B2B liquidity options

16. Matchbook: PredictStreet World Cup campaign and shared account access

17. Plaee: regulated prediction-market infrastructure partnership with CDNA

18. BitMart US: launch of its Plaee-powered prediction-market offering

19. ProphetX: Plaee integration for operator onboarding

20. Robinhood: event-contract products and exchange access

21. Robinhood: Crypto.com partnership and planned Crypto.com and OG.com equity stakes

22. DraftKings: launch of its proprietary DKeX exchange

23. DraftKings: prediction-market terms, exchanges and account arrangements

24. Underdog: launch of its UDX prediction-market exchange

25. Underdog: Prediction Picks contract rules

26. Underdog: supported exchanges and differences in customer access

27. SEC filing: Fanatics acquisition of Water Street Labs and CX Clearinghouse

28. Fanatics Markets: introducing-broker terms and CDNA relationship

29. Public: prediction-market offering and account-provider disclosures

30. Gemini: second-quarter 2026 results and Olympus clearinghouse operating update

31. Apex: proposed distribution partnership with Gemini

32. Novig: Ludlow Exchange member agreement

33. Novig: Bitnomial Clearinghouse acknowledgement

34. CFTC: Nasdaq MRX Futures exchange application

35. Nasdaq MRX Futures: proposed rulebook and CPI binary contracts

36. SEC: OCC ownership by exchange groups

37. CFTC: ForumEX exchange application

38. CFTC: Forum Clearing clearinghouse application

39. Forum Clearing: corporate ownership structure filed with the CFTC

40. ForumEX: proposed rulebook, daily settlement and contract rolling

41. Forum: attention-market product overview

42. SEC Form 4: Coinbase acquisition of The Clearing Company and merger consideration

43. CFTC: Coinbase Clearing LLC registration record

44. Coinbase: consumer prediction-market disclosure naming Kalshi

45. ICE: launch of Polymarket Signals and Sentiment data services

46. ICE: Market Signals and Sentiment service terms and trading restrictions

47. ICE: announcement of a new $600 million investment in Polymarket

48. Phantom: prediction-market routing and position disclosures

49. Polymarket: market resolution mechanisms and payout rules

Microsoft’s July 29 earnings call produced two answers to an apparently simple question: was cybersecurity mentioned? Kalshi’s final record says No while Polymarket’s says Yes.

The complete official transcript contains one occurrence of “cybersecurity,” in a question from Deutsche Bank analyst Brad Zelnick. No company speaker or operator uses that word, a plural or possessive form, or a spaced or hyphenated variant in the written transcript. Satya Nadella does use “cyber” and names containing it, which are different expressions. Microsoft transcript.

The retrieved contract descriptions draw a consequential distinction. Kalshi counts company representatives, including the operator; Polymarket counts anyone during the event. The transcript supports that explanation for the opposing results. Establishing the adjudicators’ reasoning would also require the applicable historical rules and original audio. Kalshi historical records, Polymarket final record.

The economic consequence depends on which sides someone holds. Kalshi Yes combined with Polymarket No produces no contract payout at those outcomes. Reverse the sides and both contracts win. Describing either position as “Yes on one venue and No on the other” hides the direction of the exposure.

A public earnings-market comparison had already identified Microsoft and six other candidates. This research checked five of them and found opposing final records in all five. The other two were not checked.

Walmart’s Kalshi event still carries the March 31, 2026 date from its ticker as its subtitle. Its trading dates, and its place in Kalshi’s Walmart series between the August 2025 and February 2026 calls, point to the November 20 call instead. The FedEx and Wells Fargo events also carry later ticker dates, but their subtitles name the call dates.

These were flagged cases, so the table cannot show how often such contracts disagree. No single cause is proven for all five, either. Kalshi’s summary rules require the exact word or its plural or possessive, while the four older Polymarket descriptions also count compound words. A compound such as “truckload” could therefore separate the FedEx results without any difference in speakers.

Start with the possible payouts

Suppose each contract has a maximum payout of $1. Let A be the first venue and B the second. The two possible combinations produce these gross payouts, before purchase costs and fees:

These are alternative portfolios, with potentially different entry costs. The table does not establish that either was available below $1.

If one condition is narrower than another, disagreement can be directional. A company representative is also someone on the call. Counting anyone can therefore produce Yes when counting only representatives produces No. Holding every other rule constant, that speaker distinction alone cannot produce the reverse disagreement. Event windows, phrase definitions, resolution sources, and cancellation provisions can introduce other differences.

Consider an invented purchase of 100 Kalshi Yes contracts and 100 Polymarket No contracts for $90 in total. Matching binary results return $100. The recorded Microsoft outcomes return zero, leaving a $90 loss before fees. The price is hypothetical; the payout calculation uses the recorded results.

The same logic explains why that pair can look cheap. Under the simplified speaker comparison, Polymarket’s Yes covers every case that Kalshi’s Yes covers and more, so it can fairly trade higher. Under those assumptions, Kalshi Yes plus Polymarket No can cost less than $1 because it pays nothing in the extra cases that only Polymarket counts.

A contract can settle between zero and one

Cardi B’s Super Bowl halftime appearance illustrates a partial payout. Kalshi’s contract, KXPERFORMSUPERBOWLB-26-CAR, covered a performance at the game or an announcement that she would perform. Its current record marks the strike as suspended under Kalshi Rule 13.1, with a scalar settlement of $0.26 for Yes. Polymarket’s contract asked whether she would perform live and in person during the halftime show. It resolved Yes, with a $1 payout per winning unit. Kalshi historical records, Polymarket final record.

Kalshi Yes plus Polymarket No therefore returns $0.26 per pair. The reverse position returns $1.74: $0.74 from Kalshi No and $1 from Polymarket Yes. A position can have a partial payout even when the records do not contain opposite binary answers.

The two venues reached those values by different routes. Front Office Sports reported on February 10 that Kalshi found the performance ambiguous and settled at its last-traded price, which it treated as fair value. Kalshi’s record shows that settlement late on game night, February 8, Eastern time. Polymarket’s Yes was disputed after the show, and its resolved record has a February 12 close time, also in Eastern time. The records do not show which rule versions applied to earlier trades, so the lesson here rests on the final values.

Different definitions do not always produce different results

The Hantavirus contracts inspected on October 14 illustrate a difference that exists before an outcome. Kalshi asks whether Hantavirus becomes a Public Health Emergency of International Concern in 2026, using the World Health Organization as its source. Polymarket requires an explicit WHO pandemic characterization and says a PHEIC alone is insufficient. Both contracts were open at the morning inspection. Kalshi contract, source designation, Polymarket description.

If the WHO makes the qualifying PHEIC declaration without the required pandemic characterization before the deadline, the descriptions point to Kalshi Yes and Polymarket No. Kalshi No plus Polymarket Yes then pays zero. That conditional scenario reverses the losing direction in the earnings examples; it is not a forecast about the outbreak.

The Bitcoin pair for October 14, 11:00–11:15 a.m. Eastern names different reference prices. Kalshi compares sixty-second averages of CF Benchmarks’ BRTI before the start and end of the window. Polymarket names Chainlink’s BTC/USD sixty-second TWAP stream and resolves Up when the TWAP “of the time range” is at least the price at its beginning. Kalshi listing, Polymarket listing.

Those measures can finish on different sides of their starting values. In this particular window, however, the final records agree: Kalshi No and Polymarket Down. A September Bitcoin study by Benny Ricciardi reports divergent windows elsewhere; that sample was not reproduced here. Different sources create the possibility of disagreement without ensuring that it occurs.

Refunds and half payouts differ

A refund returns what a position cost, whereas a fixed 50-cent settlement returns half the face value whatever the position cost. They can produce different cash returns even when both are casually described as cancelling a bet.

For an invented example, suppose a $1-face Yes position costs 30 cents and the opposing No position at another venue costs 60 cents. If the first venue refunds the 30-cent purchase cost while the other contract continues, the portfolio eventually returns either 30 cents or $1.30. Against its original 90-cent cost, that means a 60-cent loss or a 40-cent gain before fees. A 50-cent settlement of the first leg would instead return 50 cents or $1.50, a 40-cent loss or a 60-cent gain. In both cases the remaining contract still determines the result.

A useful comparison records the exact contract, eligible event or statistic, source, cutoff, and payout under each relevant exception. It also separates the event deadline from the time trading closes, review finishes, and funds become available.

For each proposed pairing, write down a plausible disagreement scenario and calculate both legs together. A price gap takes on a different meaning when the position can return zero, a partial amount, or a refund that leaves one leg exposed.

Scope: selected public records inspected October 14, 2026. Polymarket means its international platform. The methods and sources explain the calculations, identify the contracts, and record the evidence limits. Recorded outcomes were checked against venue APIs; customer receipts and blockchain redemptions were not audited.

1. A cancelled jobs report left two different gaps

2. The number needs an identity

3. CPI can recover a series without publishing every comparison

4. GDP changed the release identity

5. Four templates, different responses to missing data

6. A specified formula can still sit beside review powers

7. Expiration is only one part of the timing

8. Read the missing-data clause before the number goes missing

On December 16, 2025, the Bureau of Labor Statistics published the November Employment Situation report. It contained the first October payroll estimates, even though the October report itself had been cancelled during the federal shutdown. October's unemployment rate remained absent. BLS did not collect October observations for the household survey that produces the rate, either during the shutdown or afterward. BLS employment release, household-survey account.

An economic calendar could show the cancelled report and its successor without telling you which statistic had recovered. An event contract depends on whether the particular input it requires was published.

When a contract asks whether an economic number clears a threshold, a missing observation leaves the normal comparison without an input. It cannot be counted as a missed threshold. The contract's provisions for missing data determine what happens next.

As of October 9, 2026, BLS plans to omit October 2026 year-over-year CPI changes for indexes whose October 2025 values are unavailable. The release covering October 2026 is currently scheduled for November 10 at 8:30 a.m. Eastern. BLS guidance, question 16, release calendar.

The historical examples explain the publication problem. Four Polymarket US templates illustrate responses within one exchange, including an earlier period's statistic, settlement at last fair market prices, and exchange review. The comparison uses 2026 document versions observed on October 9; their applicability to a 2025 position or a particular current listing remains unverified. A separate Kalshi CPI example connects a publicly displayed outcome with a notice and a reproducible formula. The analysis is limited to these documents and the retained page evidence, without reconstructing traders' payments or estimating how common these practices are.

A cancelled jobs report left two different gaps

Payroll employment and unemployment appear together in the Employment Situation but come from different surveys. The establishment survey supplies payroll estimates while the household survey supplies the unemployment rate. Their ability to recover from a disruption need not be the same.

In 2025, September's Employment Situation moved from October 3 to November 20. The October release, originally scheduled for November 7, was cancelled. BLS later included initial October establishment estimates with the November report on December 16. Its account of the household survey confirms that no October CPS estimates are available. Revised publication calendar, December release, CPS shutdown explanation.

The later report supplied October's initial payroll estimate while leaving its unemployment rate unavailable. Following the publication calendar alone would miss that difference. Assigning zero to an unavailable observation would also misrepresent it.

The selected examples below separate a delayed publication, data recovered in another release, unavailable data, and a replacement release. “Days” measures the distance between publication dates. It does not measure when an exchange settled or when a trader could withdraw money.

• September 2025 CPI. Original date: Oct. 15. Later availability: Oct. 24, 2025. Calendar days: 9.

• September 2025 payrolls. Original date: Oct. 3. Later availability: Nov. 20, 2025. Calendar days: 48.

• October 2025 payrolls. Original date: Nov. 7. Later availability: Dec. 16, 2025, in a later report. Calendar days: 39.

• October 2025 unemployment rate. Original date: Nov. 7. Later availability: Unavailable. Calendar days: Not applicable.

• October 2025 headline CPI. Original date: Nov. 13. Later availability: Unavailable. Calendar days: Not applicable.

• November 2025 all-items CPI-U, seasonally adjusted one-month change. Original date: Dec. 10. Later availability: Unavailable in the Dec. 18 release. Calendar days: Not applicable.

• Q3 2025 GDP, advance slot. Original date: Oct. 30. Later availability: Dec. 23, 2025, replacement initial report. Calendar days: 54.

• Q3 2025 GDP, second slot. Original date: Nov. 26. Later availability: Same Dec. 23 replacement report. Calendar days: 27.

The eight rows describe selected statistics and scheduled release slots. The GDP rows share one replacement report. Sources and date arithmetic are in the methods companion.

The November CPI release makes the distinction especially clear. It arrived on December 18, eight days after its scheduled date, but Table A still lacked the seasonally adjusted one-month change for all-items CPI-U. The release instead reported a seasonally adjusted two-month increase of 0.2% from September to November and a 12-month increase of 2.7% before seasonal adjustment. Monthly figures for some components were available, but the all-items comparison remained missing when the report arrived. November CPI release, Table A.

An observation that was never collected requires attention to the contract's missing-data provisions. Depending on the applicable terms, the contract might wait for an eligible observation, reach a deadline, accept a defined substitute, or enter review.

The number needs an identity

“October inflation” leaves several questions unanswered. It can mean data published during October or data describing October. It can refer to an index level, a monthly change, or a year-over-year change. A reported figure can also belong to a different series or adjustment convention from the one a contract names.

Start with the release, which is the publication carrying the information. Then identify the statistic inside it, including its region, measure, and adjustment status. The reference period tells you which month or quarter the statistic describes. Its vintage tells you which version is eligible, such as the first official publication or a named GDP estimate. Finally, the condition specifies the comparison, including the threshold, precision, and treatment of equality.

Each part can change the answer without changing the headline. A November report can contain October observations, and a later database value can differ from the first published figure. The seasonally adjusted November 2025 unemployment rate was 4.6% in the December 16 release. On January 9, 2026, BLS's annual update of seasonal factors revised it to 4.5%. The FRED series inspected on October 9 also showed 4.5%. Initial release, January revision Table A, FRED unemployment series.

Under a hypothetical at least 4.6% condition, the first published figure meets the threshold and the revised figure does not. The payroll and unemployment templates' delayed-release clauses use the first value reported for the specified period before expiration. Their earlier-period fallback clauses leave a separate vintage question, discussed below. A data repository can show the correct latest value while the contract requires a different version.

The PM-US CPI template makes the measure and adjustment convention explicit parameters. It distinguishes the first official, non-preliminary publication from provisional figures and later revisions. Its GDP template instead allows the specific estimate to be selected in the contract terms. The unemployment template likewise identifies the particular rate, rather than accepting any unemployment measure from the same agency. CPI terms, pp. 2–3, GDP terms, pp. 2–3, unemployment terms, pp. 2–3.

The source check must identify the observation that satisfies the contract's definition.

CPI can recover a series without publishing every comparison

The CPI disruption adds another layer because a percentage change depends on more than one index value.

BLS could not collect October 2025 survey data. It used carried-forward values within its calculation process and resumed publication with November data. Although there was no October CPI news release, BLS recovered most of the nonsurvey data retrospectively and published some October component values where possible. BLS shutdown FAQ, questions 2–6.

Internal imputation and publication answer different questions. A statistical agency can need an intermediate value to continue a series while declining to publish that month's index as an observed result. A trader cannot treat the calculation process as permission to fill an unavailable official value into a contract.

A 12-month percentage change compares the current month with the same month a year earlier in the same index series:

12-month change = (current index / index 12 months earlier - 1) × 100

Both index values must use the same reference base. Changing the denominator to an adjacent month changes the period being measured. BLS calculation guide.

Consider an invented series with an October index of 309 and a September index of 308. Its one-month increase is about 0.325%, calculated as (309 / 308 - 1) × 100. The 12-month calculation also needs the previous October's index.

If the previous October were known to be 300, the 12-month increase would be 3%. If that observation is unavailable, the annual comparison remains unavailable even though the current index and monthly comparison can be calculated. These are invented values for one consistent series.

One missing month can affect later comparisons with different lengths. The unavailable October all-items index left November without its published one-month change. BLS also said it would omit December 2025 two-month changes for cities on an even-month publication schedule. For the rent panel, BLS explained that April 2026's one-month rent change would effectively use the sixth root of a 12-month price relative instead of the usual sixth root of a six-month relative. The length of the comparison changed while the one-month transformation remained. BLS shutdown FAQ, questions 7–8.

The planned omission in the November 10 release extends the same gap to annual comparisons for October 2026 data. That guidance applies to indexes whose October 2025 values are unavailable. Other series and comparison periods still need their own checks.

Why calculating a substitute does not settle the question

A missing denominator invites an estimate. You can interpolate between nearby observations, carry forward an earlier value, or use a model. The arithmetic can be internally consistent without producing the official statistic required by a contract.

BLS itself describes a geometric-mean approximation for missing CPI observations when the parties to a contract agree to use it. The approximation and any percentage changes calculated from it remain unofficial. BLS missing-data factsheet.

BLS's own internal calculations show why that distinction matters. For the commodities and services survey, it carried September prices forward into October. For rent, the carried-forward observations came from April's housing panel. To update seasonal factors, it approximated missing October index values using the geometric mean of September and November. The all-items October index remained unpublished despite these internal calculations. BLS shutdown FAQ, questions 2, 5–6, and 12.

An exchange's use of an alternative depends on the applicable contract terms and procedures as well as the calculation.

An alternative source also needs to provide the equivalent statistic. A nearby series, a different adjustment convention, or a private estimate does not become equivalent just because the number looks reasonable. The PM-US CPI template excludes ordinary third-party estimates from its normal payout criterion, while its no-data branch checks for an equivalent statistic from an alternative designated source. The other templates have different source lists, so eligibility must be read from each one. CPI terms, pp. 3–4.

GDP changed the release identity

BEA's Q3 2025 example concerns the name and identity of the publication as well as its date. The agency issued an initial estimate on December 23. Its release states that this replaced both the advance estimate scheduled for October 30 and the second estimate scheduled for November 26. BEA replacement release.

One document replaced two scheduled vintages. The 54-day interval from the advance slot and the 27-day interval from the second slot describe two paths to that same publication.

For a contract naming a particular GDP estimate, the replacement must match the required release as well as the quarter. The measure also matters, including real versus nominal GDP, annualized versus non-annualized growth, and the chosen estimate.

The PM-US GDP template makes the designated release part of the settlement input and excludes other provisional estimates and later revisions from that normal criterion. Applying it to a replacement-release scenario requires the applicable contract terms and any exchange determination about the replacement. GDP terms, pp. 2–3.

Calling the replacement “the advance estimate” for convenience would decide the identity question before reviewing the contract or the exchange's determination.

Four templates, different responses to missing data

The four PM-US documents cover the economic families in these examples: nonfarm payrolls, unemployment, CPI, and GDP. Holding the exchange constant makes their differences easier to see. Each template leaves parameters to the specific contract, and their absence-of-data clauses divide into two groups while retaining an outcome-review alternative.

• Payrolls. Ordinary delay provision: First period value reported before expiration; default expiration backstop three months after scheduled release. Missing-data or cancellation alternative: Most recent earlier period with available data, or outcome review.

• Unemployment. Ordinary delay provision: Same timing structure as payrolls. Missing-data or cancellation alternative: Most recent earlier period with available data, or outcome review.

• CPI. Ordinary delay provision: Extend expiration to actual release, unless contract terms specify otherwise. Missing-data or cancellation alternative: Last fair market prices, or outcome review; missing-data clause also checks for an equivalent statistic from a designated alternative source.

• GDP. Ordinary delay provision: Extend expiration to actual release of the specified data. Missing-data or cancellation alternative: Last fair market prices, or outcome review, if publication is cancelled or indefinitely delayed.

Condensed from the payroll, unemployment, CPI, and GDP terms. The alternatives are discretionary. Contract-specific terms and applicable modifications can change the result.

The source lists also differ. The payroll and unemployment templates name government agencies as primary sources and include Trading Economics and repositories or aggregators such as FRED as secondary sources. The CPI and GDP templates' default lists name official statistical agencies. Their authority to designate additional or substitute sources does not make every available estimate eligible. For a US CPI contract, an agency listed for another country does not by itself supply an equivalent US statistic; applicable contract terms or a designation would have to establish the eligible source. Payroll terms, p. 2, unemployment terms, pp. 2–3, CPI terms, pp. 2–4.

A repository reproducing an official series does not create its missing observation. The retained FRED unemployment series showed no October 2025 value, even while displaying November's revised figure. Source availability and vintage therefore need separate checks.

The difference in fallback mechanisms is substantive. A previous-period substitution still evaluates an economic statistic against the contract condition, but uses an earlier observation. A fair-price settlement instead requires a price determination. A review leaves the outcome to the stated process. None of these descriptions means “automatically refund the purchase price,” and none supplies a universal 50-cent payout.

For payrolls, read the fallback alongside the template's definition of its underlying: the change in total nonfarm payroll employment. The earlier-period branch therefore points to that period's reported payroll change, not its employment level or a newly calculated change across the missing months. The fallback clause does not separately specify which vintage of the earlier figure to use. That detail needs the applicable terms or an exchange determination rather than an assumption about the latest database value. Payroll terms, pp. 2–3.

A hypothetical previous-period decision

Suppose a single-month unemployment contract asks whether the specified rate is at least 4.2%. Assume its terms follow the reviewed template, the requested month's figure remains unavailable at the applicable expiration, and the most recent earlier period with data reports exactly 4.2%.

If the exchange chooses the previous-period alternative, that substituted figure meets the condition. Under the assumed $1 binary payout, a Yes contract pays $1 and a No contract pays $0. The payment in this example uses the earlier period's 4.2% rate; the missing month's rate remains unknown.

Change the otherwise identical condition to above 4.2% and the same substituted observation fails it. Yes then pays $0 and No pays $1. Equality matters after the eligible input is selected, just as it does in an ordinary release.

These outcomes are conditional examples. The exchange can choose review instead of the earlier-period branch. A relevant release before expiration, a different contract term, or an applicable modification would change the scenario. The template does not let a trader choose the favorable branch or infer the exchange's future choice.

A hypothetical fair-price decision

Now assume a different contract permits settlement “at last fair market prices” and that the price determined for Yes is $0.62. For an illustrative $1 complementary contract, that means $0.62 to Yes and $0.38 to No. The $0.62 price is stipulated; the templates do not provide a method that produces it.

A holder of 100 Yes contracts receives $62. If those contracts cost $0.40 each, the $40 purchase cost leaves a $22 gain before fees. If they cost $0.70 each, the $70 purchase cost leaves an $8 loss before fees. The settlement amount is the same in both cases because it does not reconstruct each holder's entry price.

This explains why a fair-price outcome cannot be described as a refund. The phrase does not establish a near-last-trade valuation method or show that $0.62 would be appropriate. Review remains an alternative, and no fee reversal is assumed.

A specified formula can still sit beside review powers

Kalshi's retained CPI template provides a different default mechanism from the PM-US CPI template. When data are unavailable at expiration, it calculates a substitute from the latest available CPI index and its value 12 months earlier. Its unemployment template instead specifies the last available month's data. Both documents also preserve outcome-review discretion and authority to determine payouts when an expiration value cannot be determined. Kalshi CPI terms, p. 2, unemployment terms, p. 2.

The October 2025 CPI page provides a concrete illustration. When read on October 9, 2026, its notice specified September as the fallback input month even if November data became available before expiration. October CPI page.

Using the September period named by the notice and the published unadjusted all-items indexes, the retained CPI template's formula gives:

100 × [(September 2025 index / September 2024 index)^(1/12) - 1]
= 100 × [(324.800 / 315.301)^(1/12) - 1]
≈ 0.248%

The notice required conversion to a percentage and rounding to one decimal place. Applying that rounding to the calculation gives 0.2%, matching the page's displayed outcome of 0.2.

These calculation inputs are the published all-items CPI-U indexes before seasonal adjustment, with the same 1982–84 reference base. The page notice does not identify the exact series or numerical inputs used in execution. September 2025 release, September 2024 release.

The formula produces a constant monthly rate implied by a 12-month index ratio. October's actual one-month observation remains missing. The retained template's normal underlying is a seasonally adjusted monthly change, which BLS reported as 0.3% for September. That published measure and the rate implied by the annual ratio answer different mathematical questions.

The notice's publication date, the exact governing version of the undated terms PDF, and the actual settlement and payment dates remain unverified. Together with the unidentified execution inputs, these gaps limit what we can conclude about what a holder knew before settlement or when payment occurred.

The useful contrast is the specificity of the substitute. The retained Kalshi CPI terms describe a formula, while the PM-US CPI template offers last fair market prices or review in its cancellation branch. Both preserve review powers. Reading only the common CPI headline would miss that difference in the response to unavailable data.

Expiration is only one part of the timing

Unless the payout criterion provides otherwise, the payroll and unemployment templates use a default expiration date tied to the earlier of the official release and three months after the scheduled release if no data arrive. Specific contract terms and applicable modifications also matter. CPI and GDP use a specified-date structure with delay provisions that can extend expiration. The documents generally set settlement for no later than the following day, unless outcome review begins.

If review begins, the normal settlement timetable gives way to the review process. The three-month provision therefore does not promise an unconditional date for receiving funds.

The September 30 PM-US rulebook lets designated exchange officers initiate outcome review before settlement. Rule 10.4 ties settlement to the date the review concludes and an outcome is determined. Rule 10.3 permits specified modifications, including expiration adjustments, with notification requirements. Neither provision supplies a fixed maximum review duration in the text examined here. Rulebook, rules 10.3–10.4, printed p. 80.

The document hierarchy also matters. Rule 1.5 gives product specifications priority over conflicting general rules and specific contract terms priority over conflicting product specifications. The rulebook defines a qualifying modification under Rule 10.3 as becoming part of the contract terms when issued. This makes applicable modifications part of the input to a contract comparison, alongside the listing and template. A notice should not be treated as such a modification merely because it appears on an exchange website. Rulebook, definition of Modification, printed p. 8, and rule 1.5, printed p. 13.

The agency's publication time, the contract's expiration, and the exchange's settlement process each need separate evidence. The historical table measures publication delays only; the duration of any holder's capital commitment or wait for payment remains outside its scope.

Read the missing-data clause before the number goes missing

The useful sequence begins with a precise description of the normal input. Name the statistic, geography, adjustment convention, reference period, eligible release, and comparison. Then locate the source hierarchy and the provisions for delay, missing data, expiration, and review. That makes it possible to distinguish a recoverable publication delay from an absent observation or a substituted release.

An economic forecast or calculated estimate can inform a trade while leaving the contract's required observation unavailable. Where the terms offer several fallback alternatives, the exchange's choice also remains unresolved.

Two contracts can share an economic headline and a statistical agency while requiring different inputs or specifying different responses to missing data. Comparing them requires matching both the normal observation and the exceptional settlement provisions. A calendar date alone cannot establish that they ask the same question or will pay on the same basis.

The methods companion records the source versions, clause locators, selected cases, and reproducible calculations. Historical publication facts, prospective agency guidance, and hypothetical contract examples are kept separate throughout the analysis.

Contents

1. How often did the fields differ?

2. The report contains more than one temperature

3. A maximum also needs a time window

4. Start with the contract’s report

Miami’s routine temperature readings reached 70°F on January 20, 2025, yet NOAA’s archived daily high was 73°F despite a routine reading in every hour.

That difference can matter for a temperature contract because a chart of the station’s hourly temperatures does not necessarily show the value in the contract’s named report. Polymarket US names Miami International Airport as its Miami station and the National Weather Service’s Daily Climate Report as its settlement source. PM-US weather FAQ

But “hourly data misses the high” is too broad: the same Miami archive contains a midnight report with a separate maximum-temperature remark of 73°F. The information was there in a different field.

How often did the fields differ?

To see how often the fields differed, I compared a full year of NOAA records at all five stations in the current PM-US weather FAQ: Central Park, Midway, Miami International, San Francisco International and Los Angeles International.

For each eligible date I compared the highest routine current-temperature reading with the archived daily maximum after converting both to whole degrees Fahrenheit. Each date needed readings in at least 23 distinct hours and checks for conflicting records and inconsistent timestamps.

The daily maximum exceeded the highest routine reading on 202 of 344 dates at Central Park and 186 of 352 at Midway. The corresponding counts were 241 of 350 at Miami, 234 of 351 at San Francisco and 233 of 352 at Los Angeles.

Across the stations that is about 53% to 69% of eligible dates. The usual positive difference was 1°F, though gaps of 2°F or more occurred on 6% to 15% of eligible dates depending on the station. Restricting the comparison to dates with readings in all 24 hours barely changed those percentages, so missing an entire hour does not explain the result.

These comparisons within NOAA’s historical archive are not counts of incorrect settlements. Two dates also ran in the opposite direction with the routine reading exceeding the archived high. The methods retain those cases and disclose the data checks. Data and methods

The report contains more than one temperature

A routine METAR reports weather at a particular time, with a current-temperature field that describes conditions around that observation. The daily high can occur between observations. In the Miami example the highest routine reading of 70°F came at 12:53 a.m., but the archived daily high does not record when the 73°F occurred.

Some METARs also contain maximum-temperature remarks that answer different questions from the current-temperature field. Six-hour maxima appear in reports associated with 00, 06, 12 and 18 UTC, while a separate daily maximum appears in the local-midnight report. ASOS manual, section 3.1.2

The midnight maximum remarks matched far more often in this sample. On dates with an eligible midnight maximum remark, that remark matched the archived daily high on over 99% of dates at each station. At Midway the current-temperature maximum matched on 165 of 351 dates compared with 348 of those same 351 dates for the midnight maximum remark.

Those close matches do not remove the limits of a maximum remark. The midnight report usually leaves a few minutes before the climate day ends, and the archive contains conflicting values including eight disagreements between midnight remarks and daily highs across the five stations. The remarks are useful evidence but do not replace the contract’s named report. The PM-US FAQ names the 24-hour METAR observation as a cross-check. If the CLI reading is inconsistent with it, settlement can be delayed until 11 a.m. ET for review. PM-US weather FAQ, ASOS manual, methods

A maximum also needs a time window

The Miami example shows why the window matters as well as the field. At 12:53 a.m. local standard time on January 20, a six-hour remark reported a third value of 78°F. Its window runs from about 7 p.m. on January 19 to 1 a.m. on January 20. A 6:53 p.m. reading on January 19 also showed 78°F. Assigning that entire maximum to January 20 because of the report’s date gives the wrong window.

The climate day follows local standard time throughout the year, so during daylight saving time its midnight-to-midnight period corresponds to 1 a.m. to 1 a.m. on the local clock. A midnight cutoff in a chart can therefore select the wrong period. NWS climate instruction, section 4.1.2

Start with the contract’s report

The PM-US FAQ links Miami to the NWS CLIMIA report. In the copy retrieved on October 4, the report gave a maximum of 87°F for October 3 with the date in its heading and Fahrenheit identified in its temperature table. NWS Miami CLI

The filed contract template distinguishes “above” from “at least.” Applied to 87°F, an illustrative “above 87°F” condition is false while “at least 87°F” is true. These examples explain the operators without making claims about an actual listed contract or its settlement. PM-US temperature terms

The particular contract also determines which publication counts as final. Because NWS climate reports are preliminary and can change through later quality control, a subsequently downloaded NOAA archive is not proof of the value originally used for settlement. NWS observations FAQ

For a daily temperature contract, verify the station, field, observation period, report version and comparison operator. More frequent readings help only when they measure the quantity the contract asks about.

This study measures an established reporting distinction at five stations in 2025. See Splitrule’s data and methods.

Contents

1. How the combo surcharge reaches the maker

2. Why the maker's tier matters

3. What this could mean for the trader paying the fee

4. October 4 extension: no combo maker rebate in the October 19 schedule

October 4, 2026 extension. New filings specify no combo maker rebate, with a stated October 19 start. Read the extension. The original analysis below retains its October 1 evidence and conditional results.

Updated October 1, 2026. The fee page announces a combo coefficient increase from 0.04 to 0.06 at 10 a.m. Eastern today. At 9:10 a.m., it still showed the old tables and a future notice. The calculations below use the announced curve, whose implementation we have not verified. The separate maker rebate proposal still has an intended October 6 start.

A maker supplies an order for another trader to accept. The trader who accepts it is the taker. A combo combines several outcome contracts into one position. Polymarket US's published standard maker schedule pays a $1.12 rebate on a 1,000-contract fill at 10 cents. That amount is the same for a standard contract and a combo.

At the proposed top maker rate of 15%, the standard fill would earn about $0.94. Under the announced October 1 combo curve, the combo would earn about $1.53. That rate ordinarily requires $200 million in prior-month maker volume, although other qualification routes exist. At this price, the lower maker tiers still earn less than the current formula.

The reversal comes from connecting two rules. The September 22 proposal replaces the standard maker formula with a share of taker fees. The current fee page puts combo makers on that standard formula while charging combo takers more. If the replacement covers those combo fills, some of that extra fee reaches the maker.

The proposal does not name combos, so their future coverage remains unresolved. Crypto, cross-category combos and block trades are outside this claim. The scope review records the evidence. The intended October 6 start remains subject to review. The opening amounts assume one credit rounded to cents, but the proposal does not specify actual credit rounding.

How the combo surcharge reaches the maker

Return to the 1,000-contract fill at 10 cents. The maker receives $1.12 under the current formula. The standard taker fee is $6.26. Under the announced curve, the combo taker fee becomes $10.19, up from $8.88 under the previous curve. The combo costs more because its taker fee includes a surcharge. The whole combo fee applies to the combo as a unit, not separately to each leg. These amounts follow the published formulas and October 1 notice.

That extra fee makes no difference to the maker's rebate today. Under the proposed fee share, a maker at the 15% rate receives 15% of the taker fee charged on the fill. The combo's extra $3.93 over the standard fee therefore adds roughly 59 cents to the calculated rebate. That is enough to turn the standard fill's decrease into a combo increase.

The question is how far that result extends beyond a 10-cent fill. The chart compares the proposed 15% reward with the current reward across contract prices. Its 100% line means the two formulas pay the same amount. A line above it means an increase, and a line below it means a decrease.

Maker rewards across contract prices. Splitrule calculations from the fee schedule and September 22 proposal.

The dashed line uses the standard taker coefficient of 0.0695. At the 15% maker rate, it stays at 83.4% of the current reward, a 16.6% decrease before rounding. The current maker reward and this taker fee vary with the same combination of price and quantity. Changing the price or quantity changes the dollars without changing that percentage decrease.

The combo line slopes downward because the surcharge is largest relative to the standard fee at low prices. Under the announced curve, it adds about 63% to that fee at 10 cents, versus about 11% at 50 cents before rounding. Below about 38.7 cents, the surcharge lifts the proposed top-tier reward above the old formula. Beyond that point, it still pays the maker more than a standard fill would, but less than the old reward.

Rounding near the crossing can turn a small decrease into an equal payment. Splitting an order into many small fills can also change the total. The derivation and payment examples explain why 38.7 cents is not an exact cutoff on an account statement.

The fee page read on September 30 announced a Table Tennis taker coefficient of 0.10 from 11:59 p.m. Eastern that day. That notice was absent at our October 1 read. Its removal does not establish implementation or cancellation. If the proposed 15% maker share applies there, it pays 20% more than today's 0.0125 maker formula before rounding. No source reviewed here states a separate Table Tennis maker rule. That result remains conditional, but it shows the same mechanism: higher taker fees can mean higher maker rewards under a fee share.

Why the maker's tier matters

The larger surcharge also changes the lower-tier result. At the proposed 10% maker rate, the formula can beat the current rebate below about 2.6 cents before rounding. The 5% rate still cannot. Both results assume that the proposed maker share covers these combo fills.

The second graphic returns to the same fill of 1,000 contracts at 10 cents. It changes only the assigned maker rate. Today's rebate is $1.12 in either column. The proposed amounts appear before credit rounding.

The same fill across maker tiers. Splitrule calculations from the fee schedule and September 22 proposal.

The ordinary maker tiers use the previous calendar month's maker volume. The rate starts at 5% for $10 million, rises to 10% at $50 million, and reaches 15% at $200 million. Accepted external volume can qualify a maker for a different tier. The first effective month also allows the preceding seven days of local volume scaled to a 30-day amount. The placement notes give the exact rule, and maker and taker volumes count separately.

Today's standard maker formula has no volume floor, although small fills can round to zero. Under the proposal, a maker who qualifies for no standard tier loses that rebate. At the 5% and 10% rates, the 0.0695 standard comparison pays 72.2% and 44.4% less before rounding. The 16.6% decrease is the best of those three standard-tier comparisons.

The assigned maker rate applies to the taker fees charged on that maker's fills, before taker rebates. A taker receiving more money back therefore does not reduce the maker's calculated reward on the same fill.

What this could mean for the trader paying the fee

The announced October 1 change raises the gross combo taker fee. The separate October 6 proposal would change how that fee is shared. A larger rebate could let a maker offer a better price while earning the same amount from the trade and rebate together. Competition might then pass some of that reward to the trader accepting the quote.

But the rebate alone cannot show that this happens. Makers may need the extra compensation for holding positions or trading just before prices move against them. Other liquidity rewards also affect what they can offer. This calculation measures one part of their compensation. It does not establish their profit or willingness to quote.

The taker's cost is the execution price plus the fee, less any taker rebate. The same filing reduces taker rebates in four previously rebated volume bands from $250,000 to below $50 million. Under ordinary placement, those takers keep less of their rebate. That comparison holds the fee curve fixed and comes on top of the announced combo fee increase. The methods show the four rate changes and placement limits.

A better quote helps only to the extent that it improves that total cost. The useful next question is whether quote competition offsets more of the combo surcharge where the proposed maker reward rises. This calculation points to low-priced combos at the 15% maker rate, plus a smaller region at the 10% rate. Answering that question requires execution evidence that this rule analysis does not provide.

Methods and sources

Finding decision, derivation and worked examples, source and prior-work review, claim audit and unresolved questions. The calculations use invented fills.

October 4 extension: no combo maker rebate in the October 19 schedule

Polymarket US's October 2 filings remove the condition that produced the possible combo gain in this article. Their stated start is October 19, 2026.

The combo filing sets the maker rebate to zero on all combo fills, regardless of tier. The maker fee is also zero. Under that rule, a higher combo taker fee sends no per-fill rebate to the maker. The earlier 38.7-cent crossover therefore does not apply.

For a new reader, the earlier argument was conditional: a share of a sufficiently large combo taker fee can exceed the old maker rebate. That arithmetic remains correct for the September 22 proposal and its assumed combo coverage. It does not describe the October 19 combo rule.

One invented fill under three rules. The middle amount assumes combo inclusion under the earlier proposal. The new filing excludes combos regardless of tier.

The separate maker filing changes the standard rebate to three coefficients. It no longer calculates that rebate as an exact share of the taker fee. Against the old formula, the top coefficient pays 16.8% less before rounding. For the same 1,000-contract standard fill at 10 cents, it gives $0.936, or $0.94 under the published per-fill rounding rule. Ordinary maker-volume thresholds remain $10 million, $50 million and $200 million, with placement exceptions.

The taker filing also states October 19. Its five rebate rates are 3%, 5%, 10%, 15% and 30%. Against the October 4 fee-page baseline, every previously rebated ordinary volume band receives a smaller share back. The new methods section compares all three schedules.

There is one unresolved timing question. The September 22 proposal targeted October 6, while the October 2 filings state October 19 and amend the older coefficient schedule. The reviewed public sources do not say whether the percentage proposal takes effect in between or is withdrawn. We therefore cannot establish an October 6–18 fee-share window. The October 4 fee page still lists the old maker rebate. It now describes the 0.06 combo taker curve as effective October 1. Neither statement resolves the future sequence.

The original analysis and graphics remain above as a dated explanation of the September proposal. Under the new combo exclusion, any improvement in quotes must come through other incentives or trading economics. This rule comparison does not measure execution quality.