Skip to content
DeFi Reference

Prediction markets — outcome tokens, resolution, and new market designs

1. TL;DR

Section titled “1. TL;DR”

A prediction market lets anyone trade a token that pays $1 if a specific event happens and $0 if it doesn’t, so the token’s price is a continuously-updated, money-backed estimate of the event’s probability; because a blockchain cannot natively know whether, say, a bill passed, resolution ultimately depends on an oracle — usually a bonded proposer plus a token-weighted dispute vote — which has itself become a real point of failure. Beyond simple yes/no bets, researchers have built capital-efficient designs for betting on many rare events at once, “quantum markets” that let unlimited proposals share one pool of trading capital, “multiverse finance” that turns conditional outcomes into fully composable parallel financial systems, and “opportunity markets” that let a sponsor privately crowdsource scouting without leaking its interest to competitors. Even measuring how much volume trades on these markets turns out to be non-trivial: naively summing Polymarket’s own on-chain events overstates real trading volume by roughly double.

2. Explain it simply

Section titled “2. Explain it simply”

Analogy

Section titled “Analogy”

A prediction market is like a village betting pool with a live odds board: as people place bets on whether the harvest will be good this year, the board’s displayed odds shift in real time to reflect the balance of money on each side, so anyone walking by can read off the crowd’s best current guess. When harvest time comes, a trusted village elder announces what actually happened and the pool pays out accordingly — but if two elders disagree about whether the harvest technically counted as “good,” the village has to fall back on some further process to settle the argument, which is exactly the messy part real prediction markets still have to solve.

ELI5

Section titled “ELI5”

A prediction market sells two tokens tied to a yes-or-no question: one pays $1 if the answer is “yes,” the other pays $1 if it’s “no,” and only the correct one ends up worth anything. The “yes” token’s price — between $0 and $1 — is the market’s live guess at how likely “yes” is, since people who think it’s more likely keep buying it up. The hard part is deciding who reports what actually happened: someone has to tell the program running the market the real-world outcome, and if people disagree with that report, there needs to be a fair way to settle it, usually by putting up money and letting a wider group vote.

Step-by-step walkthrough

Section titled “Step-by-step walkthrough”

Scenario: two independent, rare risk events — Event A (“a major bank fails this quarter,” 2% likely) and Event B (“a surprise rate cut this quarter,” 3% likely) — bundled into one market with three tokens: YES-A, YES-B, and NO-TO-BOTH, following a design proposed on ethresear.ch for betting cheaply on many improbable events at once.

  1. Market created (state: a writer locks $1 of collateral and mints one each of YES-A, YES-B, and NO-TO-BOTH). The three tokens are structured so their fair prices always sum to exactly $1: NO-TO-BOTH is priced at (1−a)(1−b)=0.9506(1-a)(1-b)=0.9506, YES-A at a(1−b/2)=0.0197a(1-b/2)=0.0197, and YES-B at b(1−a/2)=0.0297b(1-a/2)=0.0297.
  2. Traders take positions (state: three different traders each buy one of the three tokens at those prices). A confident trader buys NO-TO-BOTH for 95.06 cents, betting neither rare event occurs; two other traders buy YES-A and YES-B respectively as cheap insurance against their specific worry.
  3. Neither event happens (state: quarter ends uneventfully). The NO-TO-BOTH holder collects the full $1; YES-A and YES-B expire worthless — the confident trader’s 95.06-cent bet just turned into a dollar.
  4. Both events happen simultaneously — the unusual case (state: bank failure and rate cut both occur). Rather than paying out $1 to each of YES-A and YES-B (which would require $2 the writer never locked), the design “compromises by making the market have somewhat unusual behavior”: the fixed $1 of collateral is split 50/50, so each of YES-A and YES-B pays only $0.50 — the price of guaranteeing the writer’s total liability never exceeds the $1 that was actually locked up.

Common misconceptions

Section titled “Common misconceptions”
  • Myth: A prediction market’s price is a guarantee, not just a guess. Reality: It is only the market’s current, money-weighted aggregate belief — the same market design that lets a confident trader profit from being right also means that price can be, and has been, moved to reflect something other than the true probability of the underlying event (see §6).
  • Myth: Resolution is a purely automatic, on-chain process. Reality: Polymarket resolves markets through UMA’s optimistic oracle, where “anyone can propose an outcome by posting a $750 bond, anyone can challenge it within a 2-hour window,” and disputed outcomes go to “a vote of UMA token holders whose ballots are weighted by how many tokens they stake” (OddsShopper, secondary, 2025) — a human, economically-incentivized, and occasionally gameable process, not pure code.
  • Myth: “Quantum markets” involve quantum computing. Reality: The name is a physics metaphor for how the mechanism resolves multiple simultaneously-live proposals: “the quantum market triggers a ‘wave function collapse’ by observing the predicted values and selecting the proposal that predicts the highest [target metric]” (Paradigm, 2025-06) — it runs on ordinary blockchains.
  • Myth: Headline “volume” numbers from different prediction-market platforms are directly comparable. Reality: Naively summing Polymarket’s on-chain OrderFilled events double-counts both contracts and cash flow, because “there are separate OrderFilled events representing the maker side and taker side of the trade” (Slivkoff, Paradigm, 2025-12); see /exchange/prediction-market-amms/ for the full accounting mechanics.

If you only remember one thing

Section titled “If you only remember one thing”

A prediction market’s price is only as trustworthy as the process that eventually decides what “actually happened” — and that process, not the trading mechanism, is where most of the real-world drama has occurred so far.

3. How it works

Section titled “3. How it works”

Outcome tokens and splitting

Section titled “Outcome tokens and splitting”

The base primitive under nearly every prediction market is the same: $1 of collateral can be split into 1 “YES” token and 1 “NO” token (a conditional token pair), and the two can always be merged back into $1. Whichever token matches the resolved outcome redeems for $1; the other becomes worthless. The market price of the YES token is therefore a direct, tradable estimate of the event’s probability. The market-maker mechanics that make trading these tokens efficient near a coin-flip probability — LMSR, pm-AMM, and why most real volume still clears on order books rather than AMMs — are covered in full in /exchange/prediction-market-amms/; this page focuses on market design and resolution rather than that pricing math.

Resolving what actually happened

Section titled “Resolving what actually happened”

Because a blockchain has no native way to know real-world facts, resolution requires an oracle. Polymarket uses UMA’s optimistic oracle: a proposer posts a bond and states an outcome; if unchallenged within a short window it becomes final; if disputed, the question escalates to a vote of UMA token holders weighted by stake. This is economically cheap in the common case (most proposals go unchallenged) but concentrates real power in whoever can accumulate enough tokens to swing a contested vote — the mechanism ethresear.ch’s prediction-market design literature (see the reference doc below) generally treats as a separate, unsolved problem from the market’s trading mechanics.

Betting cheaply on many rare events at once

Section titled “Betting cheaply on many rare events at once”

An ethresear.ch design tackles a specific capital-efficiency problem: betting on the safe side of a lopsided market is expensive (put up $0.90 for a $1 payoff on a 90%-likely outcome), which “potentially… [leads] to prediction markets on such events systematically providing probabilities that are too far away from the extremes of 0 and 1.” For two rare, independent-ish events aa and bb, splitting the outcome space into three tokens — YES-A, YES-B, and NO-TO-BOTH — priced as

NO-TO-BOTH=(1−a)(1−b),YES-A=a(1−b2),YES-B=b(1−a2)\text{NO-TO-BOTH} = (1-a)(1-b), \qquad \text{YES-A} = a\left(1-\tfrac{b}{2}\right), \qquad \text{YES-B} = b\left(1-\tfrac{a}{2}\right)

lets a single $1 of writer collateral back protection against both events simultaneously, at the cost that if both rare events occur, “the payment is split 50/50 between the YES A and YES B sides” rather than paying out $1 to each. This generalizes to nn events via the expansion of 1−(1−x1)(1−x2)⋯(1−xn)1-(1-x_1)(1-x_2)\cdots(1-x_n): each YESi_i token claims “its fair share of all monomials containing xix_i,” and if mm events happen simultaneously, each corresponding YES token pays exactly 1/m1/m — guaranteeing the writer’s total liability never exceeds the $1 originally locked, “at a total capital lockup of $1” regardless of nn.

Beyond yes/no: quantum, multiverse, and opportunity markets

Section titled “Beyond yes/no: quantum, multiverse, and opportunity markets”

Three related 2025 Paradigm designs extend the same outcome-token primitive well past a single binary bet. Quantum markets solve the capital-fragmentation problem of decision markets with many competing proposals (e.g., “which of 700+ EIPs should Ethereum implement next?”): rather than splitting a fixed budget across every proposal, a trader deposits once and receives “an equivalent amount of tradable credits on every current and future proposal for the decision,” and at settlement the system triggers a “wave function collapse,” keeping only the winning proposal’s market live and reverting all others as if they never happened. Multiverse finance generalizes conditional tokens into fully composable “verses” — parallel universes conditioned on an event, structured with the complement/union/intersection properties of ordinary probability-theory events — so that assets can be borrowed, lent, and traded within a conditional branch (e.g., “firedETH” borrowed against “firedUSD”) without ever risking a mismatched liquidation, since both sides of a verse-conditional position vanish together the instant that verse is eliminated by resolution. Opportunity markets address a different failure: ordinary prediction markets broadcast their prices publicly, which is exactly what stops institutions (record labels, VCs, research labs) from subsidizing liquidity to scout hidden opportunities, since a public price would leak signal straight to competitors — the fix is to make prices visible only to the market’s sponsor until an “opportunity window” (e.g. two weeks) closes.

4. Worked numeric example

Section titled “4. Worked numeric example”

Continuing the two-event scenario from §2, scaled up so the numbers are easy to track: the writer locks $1,000 and mints 1,000 units each of YES-A, YES-B, and NO-TO-BOTH, priced at $950.60, $19.70, and $29.70 respectively (at $0.9506, $0.0197, and $0.0297 per token) — these three prices sum to exactly $1.00 per unit, so the $1,000 raised from selling all three token types exactly equals the $1,000 the writer must lock as collateral.

Case 1 — neither event happens. NO-TO-BOTH holders collect the full $1,000 (1,000 tokens × $1). YES-A and YES-B holders get nothing. Total payout: exactly $1,000, matching the collateral locked.

Case 2 — only Event A happens. YES-A holders collect the full $1,000. Everyone else gets nothing. Total payout: again exactly $1,000.

Case 3 — both events happen (the correlated tail case). Per the design’s explicit rule, the $1,000 is split 50/50 between YES-A and YES-B holders: each group collects $500. A YES-A buyer who paid $19.70 for what they may have assumed was “insurance paying $1,000 if a bank fails” instead receives only $500 if a rate cut happens at the same time — a real, if boundedly-sized, cost of the capital-efficiency this design buys. In no scenario does total payout exceed the $1,000 locked, which is precisely the property that lets the writer offer protection against both rare events using only as much collateral as a single ordinary binary market would require.

5. Where it’s used

Section titled “5. Where it’s used”

Ethereum

Section titled “Ethereum”
  • Polymarket — the largest crypto-native prediction market by trading volume, running on Polygon (an Ethereum-aligned proof-of-stake sidechain), using Gnosis’s Conditional Tokens Framework for outcome-token splitting/merging and UMA’s optimistic oracle for resolution; posted roughly $7.08 billion in monthly notional volume in May 2026 and roughly $10.8 billion in June 2026 (TRM Labs / DeFi Rate, secondary, dated). docs.polymarket.com
  • Kalshi — a CFTC-regulated US prediction-market exchange (not primarily a DeFi protocol, but increasingly crypto-integrated; Paradigm discloses it is an investor); closed May 2026 with $17.91 billion in monthly notional volume, its ninth consecutive monthly record, and roughly $31.5 billion in June 2026 (secondary, dated).
  • Paradigm Predictions — an analytics dashboard (not a trading venue) for exploring the broader prediction-market landscape across platforms, launched February 2026. predictions.paradigm.xyz

Solana

Section titled “Solana”
  • MetaDAO — a live futarchy (decision-market) protocol: proposals are assessed via PASS/FAIL conditional-token markets and automatically executed if PASS outperforms FAIL by a set margin; it had processed 96 proposals across 14 organizations as of mid-2026 and offers “Futarchy as a Service” adopted by Jito, Sanctum, Drift, and others (Solana Compass, secondary, as of 2026-mid). This is the live precedent Paradigm’s Quantum Markets paper explicitly cites: “to those who have heard of–or used–MetaDAO this may seem familiar.”
  • Quantum Markets reference implementation — Paradigm published a Solana/SVM reference implementation alongside its Solidity one, explicitly flagged as “reference implementations” not intended for production use (Paradigm, 2025-06).
  • Kalshi on Solana — since December 2025, Kalshi’s regulated, off-chain event contracts have been tokenized and mirrored onto Solana via DFlow and Jupiter, described as preserving “liquidity, pricing, and redemption logic” while keeping the underlying compliance framework off-chain; Kalshi added a low-latency Solana-based DoubleZero data feed in August 2026 (CoinDesk / CoinMarketCap Academy, secondary, dated).

6. Risks, attacks, and incidents

Section titled “6. Risks, attacks, and incidents”
  • Polymarket / UMA — Ukraine minerals-deal dispute, 24–25 March 2025 — $7 million paid out on a contested resolution. A market asking “Will Ukraine agree to Trump’s mineral deal before April?” moved from roughly 9% to 100% “Yes” and resolved that way despite no official agreement having been reached; a large UMA token holder cast roughly 5 million UMA tokens across three accounts — about 25% of total votes in that dispute round — to push the disputed resolution through. Polymarket called the outcome “unprecedented” and acknowledged the market had “resolved too soon,” but did not refund affected traders (OddsShopper / niftynoon, secondary, 2025-03). This is the clearest documented case of a governance attack on a prediction market’s own resolution oracle, distinct from any exploit of the trading or settlement smart contracts themselves.
  • Polymarket / UMA — MicroStrategy (Strategy) Bitcoin-sale dispute, 2026 — roughly $60–85 million at stake. Described as “the highest-dollar live test since the $237 million Zelenskyy-suit market,” this dispute turns on whether “selling in May” requires public disclosure inside the month or only on-chain execution inside the month — a genuine ambiguity in how the market’s own resolution criteria were worded, not a clear-cut factual dispute, illustrating how contract wording ambiguity can itself become an attack surface for a token-weighted vote (The Defiant, secondary, 2026).
  • Dispute volume has scaled with the market. Polymarket had “logged more than 1,150 disputed markets in 2026, already past its full-year 2025 total” as of mid-2026 (OddsShopper, secondary) — resolution disputes are not rare edge cases but a routine, scaling cost of running markets on ambiguous real-world questions at volume.
  • Volume-accounting risk is a measurement failure, not a theft, but a real integrity problem. “Most Polymarket analyses and dashboards have been mistakenly double-counting volume” by summing redundant on-chain events, inflating headline figures by roughly 2x and distorting any cross-platform or historical comparison built on that data (Slivkoff, Paradigm, 2025-12) — see /exchange/prediction-market-amms/ for the mechanics.
  • Structural risks of the newer designs. Opportunity markets are explicit about their own exploitation surface: a sponsor “has both special information about the market state at any given time, and special knowledge about their own process, which opens the risk of exploitative behavior such as hinting they will take advantage of opportunity X while aggressively selling into that market” — the paper concedes this can only be mitigated by trust and reputation, not mechanism design, at least in its initial proposed form.

7. Open problems

Section titled “7. Open problems”
  • Who governs the oracle, and how concentrated is that power? The Ukraine minerals-deal dispute shows a large token holder can, in the right conditions, outweigh the broader market’s factual consensus; no source reviewed for this page describes a fix that has actually shipped rather than being merely discussed.
  • Adverse selection in bundled multi-event designs. A later commenter on the ethresear.ch capital-efficiency thread flags a specific attack: “an attacker can exploit this by proposing an event that appears unlikely to everyone else but whose outcome the attacker secretly controls or can strongly influence,” diluting every other winning token’s payout — the thread’s proposed mitigation (a curator who screens proposed events for hidden-influence risk) reintroduces exactly the kind of trusted intermediary these designs otherwise try to avoid.
  • Whether quantum markets’ capital-efficiency actually holds up against real adversarial proposal spam. Paradigm’s own quantum markets paper flags the code as reference-only and explicitly leaves parameterization (how proposals are rate-limited, how AI-agent-submitted proposals are priced) as open future work.
  • Regulatory status remains genuinely unsettled for on-chain, crypto-native prediction markets operating in or serving the United States, an issue this page’s sources touch on only tangentially (via Kalshi’s CFTC-regulated status as a point of contrast to Polymarket) rather than resolve.

8. Ethereum vs Solana

Section titled “8. Ethereum vs Solana”
AspectEthereumSolana
Dominant live venuePolymarket (on Polygon), using CTF + UMA optimistic oracleMetaDAO (futarchy/decision markets, live production)
Primary use case observedEvent betting (politics, sports, macro) at large scaleGovernance and token-launch decision-making (“Futarchy as a Service”)
Newest research designsQuantum, multiverse, and opportunity markets (Paradigm, 2025) — Solidity + Uniswap v4 hooks reference codeQuantum markets ships an explicit SVM reference implementation alongside the Solidity one
Regulated/off-chain hybridKalshi (CFTC-regulated, US)Kalshi’s contracts tokenized and mirrored onto Solana since Dec 2025 via DFlow/Jupiter
Highest-profile documented riskGovernance capture of the UMA resolution oracle (Mar 2025, $7M)None found in sources reviewed specific to a Solana-native prediction-market exploit

Ethereum’s ecosystem hosts the largest pure-trading prediction market (Polymarket) and is where the oracle-governance risk has actually materialized at scale; Solana’s most mature related primitive, MetaDAO, is a decision-market (futarchy) rather than an event-betting market, and the newest experimental designs (quantum, multiverse, opportunity markets) are Ethereum-first in research but have been shipped with day-one Solana reference code, suggesting the gap between the two ecosystems here is more about production maturity than architectural feasibility.

9. Reference doc

Section titled “9. Reference doc”

The reference

Section titled “The reference”

Quantum Markets — Alpin Yukseloglu, Sofiane Larbi (Paradigm), 12 June 2025. paradigm.xyz/writing/quantum-markets

Summary of the reference

Section titled “Summary of the reference”

The paper opens with a capital-efficiency critique of today’s decision markets: evaluating one proposal out of, say, 700+ candidate EIPs requires a trader to split a fixed budget across every live proposal, leaving each market thin. Quantum markets solve this by decoupling deposits from proposals: a trader deposits funds once and receives an equivalent amount of tradable credit on every current and future proposal tied to a decision, with no additional capital required as new proposals appear. As proposals are traded, each accrues an implied predicted value for the decision’s target metric (e.g., ETH price); at settlement, the system observes all predicted values and “triggers a ‘wave function collapse’” by keeping only the highest-predicting proposal live and fully reverting every other proposal’s trades as if they never happened, so a trader who never touched the winning market simply gets their principal back. A worked EIP-selection example and a token-launchpad example illustrate the mechanism concretely, and the paper stresses that this makes decision markets dynamic rather than static: anyone can add a new proposal to an already-live decision at any time, which the authors argue is essential both for organic human participation and, in a longer discussion, for letting AI agents propose and evaluate options at inference-time cost rather than requiring fresh liquidity bootstrapping for every new idea. Reference implementations are provided in both Solidity (using Uniswap v4 hooks) and Solana/SVM, explicitly marked as non-production code.

Key quotes

Section titled “Key quotes”

“In the standard approach for decision markets, you would need to spread your $1M across all 700+ proposals. This would give you on average less than $1,500 per market to trade.” (Background)

“In a quantum market, traders would be able to deposit funds into the system and get an equivalent amount of tradable credits on every current and future proposal for the decision.” (Core Mechanism)

“When the settlement time is reached, the quantum market triggers a ‘wave function collapse’ by observing the predicted values and selecting the proposal that predicts the highest [value].” (Core Mechanism)

“Since quantum markets have no marginal cost of liquidity for each new proposal, AI agents can run on them at the cost of inference.” (AI Agents and Quantum Markets)

How to read the original

Section titled “How to read the original”

Background needed: the basic outcome-token/conditional-market vocabulary from §3, and familiarity with MetaDAO-style futarchy (using a market to decide between competing proposals) is genuinely helpful context, since the paper assumes some passing awareness of it. Skip the AI-agents discussion on a first pass if your interest is purely mechanism — it is a motivating use case, not part of the core design. The hardest part to internalize is the settlement step: the “wave function collapse” language is evocative but the actual mechanism is simple bookkeeping — every losing proposal’s trades are unwound as if they never occurred, which is only possible because deposits were shared credit across proposals rather than money actually spent into each one; re-read the “Status Quo Path” vs. “Quantum Market Path” comparison slowly if this doesn’t click immediately.

What changed since

Section titled “What changed since”
  • Paradigm’s own follow-up papers the same year — Multiverse Finance (2025-05) and Opportunity Markets (2025-08) — extend the same outcome-token substrate to conditional-composability and private-liquidity problems respectively (see §3).
  • MetaDAO, the live Solana futarchy protocol this paper cites as prior art, continued scaling in production through 2026 (96 proposals across 14 organizations as of mid-2026, per secondary reporting), offering an empirical comparison point for how static decision markets perform at real scale even without the quantum-markets capital-efficiency fix.
  • As of this page’s writing (2026-08), no production deployment of the quantum-markets mechanism itself (as opposed to its reference code) was found in the sources reviewed.

Secondary references

Section titled “Secondary references”
  1. “Multiverse Finance” (Paradigm, 2025-05) — read if you want conditional-token composability generalized beyond a single decision’s proposals.
  2. “Opportunity Markets” (Paradigm, 2025-08) — read if you want the private-liquidity, sponsor-subsidized variant of the same outcome-token primitive.
  3. MetaDAO documentation — read for the live, production futarchy precedent this paper builds on conceptually.

The reference

Section titled “The reference”

Prediction market design for betting on many highly improbable events — Vitalik Buterin, ethresear.ch, 29 November 2020. ethresear.ch/t/8280

Summary of the reference

Section titled “Summary of the reference”

The post identifies a specific capital-efficiency failure in ordinary binary prediction markets: betting on a highly likely outcome costs nearly $1 per $1 of eventual payoff, which the author argues systematically discourages confident “nothing unusual will happen” bets and leaves rare-event markets mispriced away from the true extremes. The proposed fix, worked out first for two events aa and bb, splits the outcome space into three tokens — YES-A, YES-B, and NO-TO-BOTH — priced so their sum is always exactly 1, letting someone take a cheap position against each rare event while a single writer’s $1 of collateral backs all three. The design is explicit about its trade-off: if multiple rare events occur simultaneously, the fixed collateral is split proportionally among the corresponding YES tokens rather than paying each in full, so “there is no way to win $N on all N events happening at the same time.” The post generalizes the two-event construction to nn events via the algebraic expansion of 1−∏i(1−xi)1-\prod_i(1-x_i) (subtracting the product of every event’s complement from one), and to other multi-outcome event structures like Augur’s “INVALID” category. A long reply thread — spanning 2020 through 2026 — debates whether the added complexity should sit on the “YES” side (as the original post does) or be shifted to specialized “NO”-side liquidators who absorb it for a fee, connects the design explicitly to real-world sports “parlays,” proposes an LMSR-based implementation of the same idea, and, in the thread’s most recent (2026) reply, identifies a concrete adverse-selection attack: an attacker can propose a plausible-looking rare event whose outcome they secretly control, buy the corresponding cheap YES token, force it to resolve true, and dilute every other winning token’s payout — a risk the original 2020 design does not address.

Key quotes

Section titled “Key quotes”

“If the probability of some event is 90%, someone wishing to bet for that event must put up $0.9 of capital per $1 of position, whereas someone betting against the event need only put up $0.1 of capital.” (Introduction)

“The design allows taking a $1 position against each of the N improbable events at a total capital lockup of $1.” (Introduction)

“If only one event xi happens, the holder of the YES xi token gets a full $1, but if m events xi, xj … xz all happen, then the holder of each corresponding YES token gets paid $1/m.” (Expanding to more than two assets)

“An attacker can exploit this by proposing an event that appears unlikely to everyone else but whose outcome the attacker secretly controls or can strongly influence.” (Reply #24, distbit0, 2026-08)

How to read the original

Section titled “How to read the original”

Background needed: basic probability of independent events (products and complements) and the conditional-token split/merge primitive from §3. Skip the “emergently discovering which side of an event is improbable” extension paragraph on a first pass — it is flagged by the author as future work, not a developed part of the mechanism. The hardest part is the pricing formulas for n>2n>2 events: rather than trying to memorize the general monomial-expansion formula, work through the two-event case by hand first (verify (1−a)(1−b)+a(1−b/2)+b(1−a/2)=1(1-a)(1-b) + a(1-b/2) + b(1-a/2) = 1 algebraically) and only then trust that the same fair-share logic extends to more events.

What changed since

Section titled “What changed since”
  • The 2026-08 reply identifying the adverse-selection/curation problem shows the design’s core vulnerability was only fully articulated six years after the original 2020 proposal, during a period (per the reply timestamps) of renewed practical interest in this mechanism.
  • Paradigm’s 2025-06 Quantum Markets design solves a related but distinct capital-efficiency problem (many competing proposals for one decision, rather than many independent rare events) using a different mechanism (shared tradable credit rather than a fixed-collateral multi-token split).
  • No production implementation of this specific many-improbable-events design was found in the sources reviewed for this page as of 2026-08.

Secondary references

Section titled “Secondary references”
  1. “Continuous Prediction Markets” (ethresear.ch, 2022-07) — read if you want a related mechanism-design thread on evolving market structures over time rather than a fixed multi-event bundle.
  2. LMSR background (Hanson) — read for the automated-market-maker alternative implementation proposed in reply #16 of this thread, and see /exchange/prediction-market-amms/ for the fuller LMSR/pm-AMM treatment.
  3. Polymarket resolution documentation (docs.polymarket.com) — read for how a live, large-scale platform actually implements dispute resolution, as distinct from this post’s pure market-design focus.

10. Sources

Section titled “10. Sources”