Key takeaways
- Conditional markets are built on a pricing identity that assumes one shared event. In its October 2026 survey of conditional markets, Galaxy Research describes how a prediction market's price for an outcome (the "state price") can be combined with the price of an asset conditioned on that outcome. The combination recovers the event's expected impact on the asset. The identity holds only if both markets define the event in the same way.
- The identity magnifies definitional error. The impact statistic's sensitivity to the state price is 1 / [π(1 − π)], where π is the state price, so a small mismatch in the definition becomes a large error in the estimate. In Galaxy's own orange-juice example, the state price is 10%. Shifting it to 14%, the kind of gap two different definitions of "a freeze" could produce, cuts the estimated impact by more than half. At 20%, the sign flips.
- Real outcome spaces are not clean binary partitions. In our sample of high-volume Polymarket International markets, one resolved market in five carried an equal-split fallback. Kalshi settles leader-out contracts at a scalar value when the leader dies. Polymarket US settles contracts on natural persons at last traded prices under its Rule 10.5. A conditional asset built on any of these predicates needs a rule for the extra states, and a prediction-market price that includes them is not a pure state price.
- Event identity is fragmented even inside single venues. 593 of Kalshi's 15,579 series (3.8%) carry settlement-source metadata that reads as sentence fragments, and some list sources unrelated to the contract. In our Polymarket sample, 11.1% of markets have no resolution-source field at all. The binding terms live in prose. Meanwhile, US exchanges listed about 1,600 new event contracts in 2025, against an average of five a year between 2006 and 2020.
- Finance has solved this class of problem before by separating defining an event from trading it. Precedents include ISDA's credit-event definitions and Determinations Committees, the 2019 rule change after a manufactured default, regulated benchmarks and fallback protocols. Event markets need the equivalent: a versioned, venue-neutral event record with eight core fields, plus typed relationships to each venue's contract. We set out what that record must contain and who could maintain it.
1. Where Galaxy's map leads
In "Conditional Markets Are Beginning to Blossom" (15 October 2026), Galaxy Research surveys the generation of event-driven markets that followed prediction markets.1 Two types are converging into one category:
- Decision markets settle conditional trades and also bind an organization to the outcome that prices favor. Galaxy counts $50.5 million in treasuries governed this way across 24 organizations.
- Impact markets price an asset in the world where an event happens and in the world where it doesn't. The gap between the two prices measures the event's expected effect on the asset.
Galaxy also covers infrastructure. Branchpoint's Branchmap is a proposed onchain namespace that gives "events, predicates, and outcomes stable, machine-readable identities". A prediction market, a DEX and a lending protocol could then all reference one event without each defining it separately.1 Galaxy also discusses the "statebook" proposal, which assembles claims on states of the world from existing instruments. It concludes that these designs share one idea: markets organized around the states of the world traders want exposure to.
We agree with the direction, and this paper adds a constraint the map implies but doesn't measure. A state of the world is only as useful as its definition. Today, different venues define the "same" state differently, so markets that combine them inherit the difference. Galaxy notes that Branchmap's usefulness depends on venues adopting a shared substrate instead of building isolated event definitions. Our evidence suggests venues will keep building isolated definitions for some time. The rest of this paper explains why, and what to do about it.
2. The pricing identity needs one definition
Galaxy's worked example uses a freeze. Suppose a prediction market prices "Florida temperatures fall below 28°F for more than three consecutive hours in December" at 10%. A claim on frozen concentrated orange juice that pays only in the freeze world trades at $0.40 per pound. The matching claim for the no-freeze world trades at $1.80.
Dividing each branch price by its state price gives the asset's value inside each world: $4.00 if there is a freeze and $2.00 if there isn't. The market therefore expects juice to be worth twice as much after a freeze. The impact statistic, the log of that ratio, is ln 2 ≈ 0.69.1
Written generally, with branch prices B_yes and B_no and state price π:
impact = ln(B_yes / B_no) + ln[(1 − π) / π]
The first term comes from the conditional market and the second from the prediction market. The derivative of impact with respect to π is −1 / [π(1 − π)]. That is about −11 at π = 10% and −4 at π = 50%. An error of one percentage point in the state price moves the impact estimate by about 0.11 at a 10% state price, which is a sixth of Galaxy's entire estimated effect.
Errors like that arise whenever the prediction market and the conditional market do not define the event in exactly the same way. "Below 28°F for more than three consecutive hours" is a different event from "any hourly reading below 28°F", and both differ again by weather station and time zone. Suppose the conditional market's verses are defined by a predicate whose true probability is 14%, while the analyst divides by the 10% price of a slightly different predicate on another venue.
| State price used (π) | In-freeze value (B_yes / π) | No-freeze value (B_no / (1 − π)) | Implied ratio | Impact (log ratio) |
|---|---|---|---|---|
| 8% | $5.00 | $1.96 | 2.56× | 0.94 |
| 10% (Galaxy's example) | $4.00 | $2.00 | 2.00× | 0.69 |
| 12% | $3.33 | $2.05 | 1.63× | 0.49 |
| 14% | $2.86 | $2.09 | 1.37× | 0.31 |
| 20% | $2.00 | $2.25 | 0.89× | −0.12 |
Branch prices are held at Galaxy's illustrative $0.40 (freeze) and $1.80 (no freeze). The values are arithmetic on that example, not market observations.
A four-point gap between definitions more than halves the estimated impact, and a ten-point gap flips its sign. Neither gap is unusual. Our companion paper documents same-question contracts that settled on opposite sides of the same day: a shutdown contract that paid YES on Kalshi and NO on Polymarket on 14 February 2026.2 In those cases the gap was not a few points. It was the whole outcome.
The problem is worst where conditional markets promise the most. Their most valuable use is pricing the impact of rare events: freezes, defaults, policy surprises. When π is small, 1 / [π(1 − π)] is large, so the identity is most sensitive to definitional error exactly where it is most useful.
3. The third state: outcome spaces are not clean partitions
Branchmap's mechanism, as Galaxy describes it, splits an asset into copies conditioned on each outcome of a predicate. When the predicate resolves, the surviving branch recombines into the plain asset. That assumes the outcomes partition the world: exactly one branch survives.
Live event contracts often have states that break this assumption.
| Mechanism | Where | What holders receive | Observed frequency |
|---|---|---|---|
| Equal split ("50/50") | Polymarket International | $0.50 per share on each side | 20.0% of resolved markets in our sample carried the fallback flag; 1.18% actually resolved 50/50 |
| Scalar settlement on death | Kalshi (WLEADEROUT terms; rulebook amendment of 2 March 2026) | The last traded price before death: for example, $0.02 on the YES side of "Khamenei out before March 1" | Applied to all four open Khamenei contracts on 28 February 2026 |
| Last traded price for natural persons | Polymarket US (Rule 10.5, rulebook dated 30 September 2026) | The last traded price before death or incapacity, or before those circumstances became known | Rule in force; no case observed in our data |
| Committee allocation | Kalshi Outcome Review Committee; Polymarket US outcome review | A "fair allocation" or determination set by the exchange | Case by case |
Sources: Polymarket Gamma API (sample described in the companion paper); Kalshi contract terms and Trade API v2 market records; Polymarket US Rulebook.
Two consequences follow for conditional-market design.
First, a prediction-market price is a pure state price only if the fallback probability is zero. Call the probability of a fallback φ, and suppose the YES side receives a share w of the payout in that case. Ignoring discounting, the YES price is approximately π + wφ. Suppose there is a 2% chance of an equal split (φ = 0.02, w = 0.5) on a predicate whose true YES probability is 10%. The YES price is then about 11%. Divide Galaxy's branch prices by 11% instead of 10%, and the estimated impact falls from 0.69 to 0.59. That understates the true impact by 15%, from a fallback most traders never consider.
Second, the conditional asset needs its own fallback rule. If the predicate resolves 50/50, it is not clear what each verse-conditioned claim should pay. Each might receive half the asset, or the position might unwind at the pre-event price, or the market might void. If the predicate settles at a scalar of $0.02, as Kalshi's Khamenei contract did, the "YES world" has not happened at 2%. The rule has substituted a price for a state. Any of these answers can be defensible. What is not defensible is leaving the question unanswered, because the answer determines the payoff.
4. Identity is already fragmented, even inside venues
If venues defined events consistently within their own catalogs, a shared layer would mainly need to translate between them. The evidence suggests they don't.
The machine-readable metadata is not the contract. Kalshi's public API attaches a list of settlement sources to each of its 15,579 series. On 31 October 2026:
- Fragments. 593 series (3.8%) carried source names that read as fragments of sentences, such as "The Source Agencies are" or "in hierarchical order", instead of named sources.
- Unrelated sources. Some series listed sources with no visible connection to the contract. The series for new US sanctions actions on Iran lists two Bureau of Labor Statistics releases. A grocery-inflation series lists the Federal Reserve Bank of New York, linked to its SOFR reference-rate page. A series on whether a video game sets a revenue record lists the San Francisco Unified School District.3
These look like template artifacts. Kalshi's own contract terms state that instructions on how to access the underlying are not part of the binding terms, and the contract filing and market rules govern.4 That is the point. A system that reads the structured fields instead of the binding text will sometimes be wrong about what a contract settles on.
The same holds on Polymarket International. In our sample of 96,605 markets, 11.1% had an empty resolution-source field. Their sources appear only in the prose description, and any later "additional context" is published separately.2
The brand is not the venue. Polymarket International resolves through UMA's optimistic oracle. Polymarket US resolves through exchange officers under a CFTC-filed rulebook. Hyperliquid's outcome markets, Galaxy Research reported in June, use a third model: validators publish and settle curated markets.5 Polymarket's own up-or-down crypto markets use a fourth: a deterministic Chainlink price feed.6 One question can be listed under all four.
The catalog is growing faster than anyone can read it. According to the Congressional Research Service, US exchanges listed an average of five new event contracts a year between 2006 and 2020, 131 in 2021, and about 1,600 in 2025.7 Each listing is a new contract text. One Polymarket question can spawn dozens of dated variants. In our sample, 2,100 events contained 96,605 markets.
Venues compete on contract design, and regulation fixes each venue's definitions in place. A regulated exchange self-certifies its contract terms to the CFTC, and amends them through further filings. Kalshi changed how it settles contracts whose subject dies within two days of the Khamenei case, by filing a rulebook amendment.8 An exchange cannot simply defer its definitions to an external namespace. Adopting one would require new filings, and the exchange would remain responsible for the result. That is why we expect isolated definitions to persist even as shared namespaces appear.
5. What finance learned the first time
Event markets are not the first to discover that trading a contingency requires agreeing on what the contingency is. Four precedents carry over directly.
Credit derivatives: definitions plus a standing body that decides. A credit default swap pays when a "credit event" occurs. Whether one has occurred is often contested: what counts as a restructuring, or as a failure to pay? In 2009, ISDA's "Big Bang" Protocol built auction settlement into standard contracts. It also created regional Credit Derivatives Determinations Committees to decide credit events across the market. Decisions require an 80% supermajority, and questions that fall short go to external review. More than 2,000 parties adhered within weeks.9 The design separated deciding whether the event happened from trading the contracts that depend on it, and it published the decisions.
Definitions get gamed, then rewritten. In 2017–18, a financing arrangement with Hovnanian Enterprises was structured around a deliberate missed payment intended to trigger CDS. After litigation, the payment was ultimately made. In April 2018, the CFTC warned that such "manufactured credit events" may constitute market manipulation. ISDA responded with the 2019 Narrowly Tailored Credit Event Supplement, effective 27 January 2020. It adds a credit-deterioration requirement to the failure-to-pay definition, and a parallel protocol let existing trades adopt it.10 Lesson: definitions need versions, and migrating open positions from one version to the next is itself a design problem. Event markets met the same problem in 2026, when contract wording that had been read one way was clarified, amended or codified after the fact.
Benchmarks: the fixing is a governance choice. After the LIBOR scandals, IOSCO's 2013 Principles for Financial Benchmarks and the EU and UK Benchmarks Regulations set standards for how a reference price is produced, governed and disclosed. Event markets already show the difference. Kalshi's bitcoin contracts settle on CF Benchmarks' BRTI, which the administrator describes as a registered benchmark under the UK Benchmarks Regulation. Polymarket International's settle on a single exchange's one-minute candle.11 Both are legitimate sources, governed very differently.
Fallbacks: decide what happens before the source disappears. ISDA's 2020 IBOR Fallbacks Protocol, effective January 2021, gave derivatives referencing disappearing interbank rates a predefined replacement. Event contracts face smaller versions of the same problem all the time: delayed statistical releases, cancelled meetings, postponed games, and deaths.
Identifiers name instruments, not events. ISINs, CFI codes, FIGIs and the OTC derivatives Unique Product Identifier name instruments and product types. None of them identifies the real-world proposition two venues' contracts both reference. Two contracts on the same election get unrelated identifiers. No current identifier links them as the same question.
6. A minimum event record
The precedents point to the same design: a venue-neutral record of the event, separate from any venue's contract, with each contract linked to it by an explicit, reviewed relationship. We propose eight fields as the minimum.
| # | Field | What it captures | What goes wrong without it |
|---|---|---|---|
| 1 | Proposition | The real-world claim, worded neutrally and independent of any venue's title | Contracts that sound alike get treated as the same; Khamenei "out" and "remains" were framed as opposites |
| 2 | Outcome space | Every outcome, including fallback states (equal split, scalar, void) and how each one pays | Exhibit 2; conditional assets with no rule for the third state |
| 3 | Trigger and evidentiary standard | Whether the contract is event-, observation-, announcement- or confirmation-based; what evidence counts; who must provide it, and by when | 14 February shutdown (observation vs announcement); Strategy (event vs confirmation) |
| 4 | Source hierarchy | Primary source, ordered fallbacks, and the conditions under which the venue may substitute a source | Kalshi FOMC terms permit source substitution; metadata that lists the wrong source |
| 5 | Fixing and time semantics | Point or window, time zone, inclusive or exclusive thresholds, unit and quote currency | Bitcoin on 14 October 2026: BRTI 60-second average at 5 p.m. vs a Binance BTC/USDT candle at noon |
| 6 | Contingencies | Death, incapacity, cancellation, rescheduling, missing data, rounding | FOMC cancellation and rounding clauses; natural-person rules |
| 7 | Adjudication | Who decides, how disputes are handled, whether terms can be clarified or modified, and finality | Exchange officers vs token-holder vote; disputes add days |
| 8 | Version and provenance | The terms text as captured, its hash and capture time, effective dates, and later clarifications | Clarifications that change the effective reading after trading has begun |
The record only becomes useful through its relationships. Each venue contract should link to the canonical event with a typed relationship and a review state. The relationship might be exact equivalence, a nested or overlapping condition, a complement, a member of a mutually exclusive set, or only the same topic.
Predict Here's public methodology implements one version of this model:
- relationship types including exact complements, mutually exclusive sets, exhaustive sets and cross-venue equivalence;
- four semantic states (VERIFIED, PARTIAL, NOT_EQUIVALENT and UNREVIEWED), with nothing treated as equivalent by default;
- a rule that only verified relationships can feed an aggregate figure.12
The specific vocabulary matters less than three principles:
- Equivalence is a reviewed claim, not a text match.
- "Not comparable" is a valid answer.
- Every claim carries the version of the terms it was made against.
7. Who should maintain it?
There are five candidates, and they are not mutually exclusive.
| Model | Strength | Limitation |
|---|---|---|
| Each venue alone (today) | Legally authoritative for its own contracts | No identity across venues; definitions drift independently |
| Shared onchain namespace (for example, Branchmap) | Composable across onchain protocols; references are machine-readable | Depends on venue adoption; regulated exchanges' binding terms live in CFTC filings, not in a namespace |
| Industry determinations body (an ISDA-style committee) | Can make binding decisions for contracts that incorporate it; decisions are published | Slow to form; needs participation from venues and major traders; best suited to high-value event families |
| Independent reference layer | Neutral across venues; reads binding terms; records relationships, versions and review states; fast to extend | Informs users but cannot bind venues; its quality depends on review discipline |
| Regulator | Sets minimum standards for regulated venues (the CFTC's June 2026 proposal names "settlement integrity deficits" as a public-interest factor) | No reach offshore; sets minimums, not shared definitions |
Our view. In the near term, the work falls to independent reference layers and to venues making their own terms more explicit. Users need cross-venue identity now, and no other model can deliver it soon. For onchain conditional markets, namespaces such as Branchmap are the natural home, and a reference layer should map to them where they exist. If institutional activity in event contracts keeps growing, the highest-value event families are good candidates for something closer to a determinations committee. These include elections, leadership changes, central-bank decisions and government funding: the questions where the companion paper found disputes concentrated. Determinations would be published, and venues and participants could choose to incorporate them by reference.
8. Our take
Identity comes before liquidity. Galaxy's survey ends on liquidity: conditional instruments will have to build it largely from scratch. We would add an earlier step. Liquidity in conditional markets will come partly from arbitrage against prediction markets and perpetual futures. That arbitrage is only as clean as the event definitions it bridges. If the definition is wrong, the arbitrage becomes basis risk.
The rarest events are the most fragile. The impact identity is most sensitive to definitional error when the state price is small. Freezes, defaults and policy shocks are where conditional markets are most valuable, and where a shared definition matters most.
Design for the third state. Every conditional-market design should state what happens when its predicate resolves 50/50, settles at a scalar value, or voids. Every price identity should account for the probability of that happening.
Expect isolated definitions to persist, and build bridges. Regulated venues self-certify their own terms and compete on contract design. Offshore venues resolve ambiguity after the fact. A shared namespace is a good destination, but it will not arrive on its own. Until it does, the industry needs a venue-neutral record of what each contract actually pays on, maintained by people who read the terms.
Predict Here keeps that record: canonical events kept separate from venue contracts, reviewed relationships with explicit semantic states, and versioned terms with source and capture time. Read our standards, governance and methodology for the current model and its limits.
Methodology and limitations
Exhibits 1 and 2.
- Exhibit 1 is arithmetic on Galaxy Research's published illustrative example. It is not a market observation.
- Exhibit 2's frequencies come from the Polymarket sample described in our companion paper: 96,414 resolved markets in the 2,100 highest-volume closed events with end dates on or after 1 October 2025, retrieved 31 October 2026. They also draw on Kalshi's public market records.
Kalshi metadata.
- Counts come from Kalshi's public
/seriesendpoint, retrieved on 31 October 2026. - "Fragment-like" names are source names matching patterns such as "The Source Agencies are", "in hierarchical order", or names that end mid-phrase.
- The specific examples cited are KXOFAC, KXGROCERYCPI and KXGTASALESRECORD.
- These counts describe metadata only. They do not show that any contract settled incorrectly.
Sources. Descriptions of ISDA, IOSCO and benchmark arrangements are summaries for context, drawn from public sources and law-firm briefings. They are not legal analysis.
Notes and references
- Galaxy Research, "Conditional Markets Are Beginning to Blossom" (15 October 2026). https://www.galaxy.com/insights/research/conditional-markets-decision-markets-impact-markets-mainnet-umia-metadao-proof-branchmap-branchpoint-prediction-markets-futarchy ↩
- Predict Here Research, "Same Question, Different Contract" (companion paper, 31 October 2026). https://predicthere.com/insights/same-question-different-contract ↩
- Kalshi Trade API v2,
/series(settlement_sources), retrieved 31 October 2026. Series cited: KXOFAC ("New Iran OFAC action by [date]"), KXGROCERYCPI ("Grocery CPI YoY"), KXGTASALESRECORD ("Will GTA 6 set the all-time 24-hour gaming revenue record?"). ↩ - KalshiEX LLC, FEDDECISION contract terms ("These instructions on how to access the Underlying are provided for convenience only and are not part of the binding Terms and Conditions"). https://assets.kalshi.com/regulatory/product-certifications/FEDDECISION.pdf ↩
- Galaxy Research, "The Race to Trade Everything" (9 June 2026). https://www.galaxy.com/insights/research/hyperliquid-hip-4-prediction-markets-polymarket-kalshi-analysis ↩
- Polymarket documentation, "How Resolution Works." https://docs.polymarket.com/concepts/resolution ↩
- Congressional Research Service, Legal Sidebar LSB11441 (2026). https://www.everycrsreport.com/reports/LSB11441.html ↩
- DeFi Rate, "Kalshi Codifies Death Settlement Rule in CFTC Filing Amid Iran Market Backlash" (March 2026). ↩
- ISDA, "Big Bang Protocol" (March 2009) and announcements on the implementation of the Determinations Committees; Paul, Weiss and Mondaq summaries of the Protocol's DC Rules (80% supermajority; external review). ↩
- Clifford Chance, "Narrowly Tailored Credit Event" briefing (February 2020); Simmons & Simmons, "ISDA publishes Narrowly Tailored Credit Event Protocol" (2019); CFTC staff statement on manufactured credit events (April 2018), as summarized therein. ↩
- CF Benchmarks, BRTI page and FCA authorization notice. https://www.cfbenchmarks.com/indices/BRTI ↩
- Predict Here, Methodology and Governance. https://predicthere.com/methodology; https://predicthere.com/governance ↩
Disclosures
This paper is general research published for information only. It is not investment, legal or tax advice, and it does not recommend any position, venue or strategy. References to Galaxy Research, Branchpoint and other organizations describe their published work. They do not imply any endorsement of, or relationship with, Predict Here. Predict Here is operated by Odyssea Research LLC. It does not list markets, hold client funds, take the other side of trades, or execute or route orders.
Persons associated with Predict Here may hold positions in event contracts, including on venues discussed in this report.