All posts

Prediction market development: how to build one in 2026

How to build a prediction market in 2026: CLOB with conditional tokens, LMSR and parimutuel designs, UMA and other oracles, resolution disputes, US and EU regulation, and cost by scope.

By , CEO & Founder at 23studio

Published 11 min read

Prediction market development: how to build one in 2026

To build a prediction market, you pick a trading mechanism (an order book over outcome tokens like Polymarket, an LMSR market maker, or a parimutuel pool), mint outcome tokens against collateral, connect an oracle that resolves each market, and wrap it in a frontend, indexer and market-creation workflow. An on-chain MVP takes about 2 to 4 months. The legal structure often takes longer than the code.

This guide covers the three market mechanisms and when each fits, how resolution and oracle disputes work in practice, the components you need, the US and EU regulatory picture as of late 2026, and cost and timeline ranges by scope.

How prediction markets work

A prediction market sells shares that pay out if an event happens. A "Yes" share on a binary market pays $1 if the outcome is Yes and $0 otherwise, so its price reads as the market's probability. The platform has three jobs: let people trade those shares at fair prices, hold the collateral so payouts are guaranteed, and decide what happened.

On-chain, the cleanest way to guarantee payouts is full collateralization. One dollar of collateral mints one Yes and one No share. Together they're always worth exactly one dollar, and after resolution the winning share redeems for the dollar. No one is ever owed money the system doesn't hold.

Market mechanisms: CLOB, LMSR or parimutuel

MechanismHow prices formLiquidity needsBest forUsed by
Order book over conditional tokensTraders post bids and asks on outcome tokensMarket makers per marketHigh-volume markets with professional tradersPolymarket, Kalshi (centralized)
LMSR automated market makerA cost function quotes prices from outstanding sharesA subsidy per market, bounded lossMany small or long-tail markets, internal forecastingEarly Gnosis and Augur-era designs, many research markets
Parimutuel poolAll stakes pooled, winners split the pool pro rataNone up frontSimple bets, sports-style pools, social appsMany betting-style dApps

Order book with conditional tokens

This is the Polymarket model, and what most people mean by "Polymarket clone". It has two layers.

The token layer is the Gnosis Conditional Tokens Framework (CTF). It lets anyone split collateral into ERC-1155 outcome tokens for a condition, merge a full set back into collateral, and redeem winning tokens after an oracle reports the result. It also supports combinatorial positions, such as "Team A wins and turnout is above 60%".

The trading layer is an order book. Polymarket's original CTF Exchange describes itself as "a hybrid-decentralized exchange model wherein there is an operator that provides offchain matching services", with "settlement on-chain, non-custodially". Users sign orders, the operator matches them, and the contract settles the swap of outcome tokens and collateral. If a buyer of Yes and a buyer of No match, the exchange can mint a new pair from their combined collateral, so a market doesn't need pre-minted inventory.

Polymarket is now moving past the Gnosis CTF. Its Protocol V2, introduced in October 2026, replaces CTF with a single ERC-1155 position contract, uses a dollar wrapper called pUSD as the only collateral, and routes resolution through an oracle aggregator that combines UMA, Chainlink Data Streams for price-based markets and manual reporting, according to Parameter's coverage. New markets were scheduled to move to V2 from November 2, 2026, with existing CTF markets left in place. For a new build, the CTF remains a solid, well-understood base, and Polymarket's V2 design shows where the pattern is heading.

LMSR

Robin Hanson's logarithmic market scoring rule is an automated market maker built for prediction markets. A cost function prices every trade from the number of outstanding shares in each outcome. The operator sets a liquidity parameter b and the market maker's worst-case loss is bounded at b · ln(n) for n outcomes. That bounded subsidy is the whole point: you can open thousands of markets and always quote a price, without recruiting a market maker for each.

The cost is that the subsidy is real money, and prices move a lot on large trades if b is small. LMSR fits long-tail markets, internal corporate forecasting, and launching a platform before market makers show up. Some platforms start markets on an AMM and move them to an order book once volume arrives.

Parimutuel

Everyone stakes into outcome pools. When the market resolves, the winning pool splits the total, minus fees. There's no price discovery during the market beyond the pool ratios, and early bettors don't know their final odds. It's the simplest design to build and audit, and it suits casual products, mini apps and social betting. It's a poor fit for traders who want to exit before resolution.

Resolution and oracles

Resolution is where prediction markets succeed or lose user trust. Code can be perfect and the platform still fails if users believe a market resolved wrongly.

UMA's optimistic oracle

Polymarket made UMA's optimistic oracle the default for crypto prediction markets. The flow:

  1. A market request sets the resolution criteria, a proposer reward, a bond and a challenge period.
  2. A proposer posts an answer with a bond. If no one disputes during the challenge period, the answer stands and the proposer gets the reward and bond back.
  3. Anyone can dispute by posting a matching bond. Disputes escalate to UMA's Data Verification Mechanism, where staked UMA holders vote in a 24-hour commit and 24-hour reveal round. A result needs 65% of staked UMA behind one outcome.
  4. The losing side's bond pays the winner.

It's cheap and fast when no one disputes, which is most of the time. Polymarket has said only about 0.2% of its markets trigger a UMA vote. The weak point is the dispute layer: token-holder votes can be swayed by holders who also have positions in the market. After repeated disputes, UMA moved Polymarket to a managed version in August 2025 where only whitelisted addresses can propose answers, while anyone can still dispute. Bloomberg later reported that a handful of large wallets controlled about half of the voting power used in Polymarket disputes, and arbitration counts rose in 2026.

Other options

Resolution methodSpeedTrust modelGood for
Price feeds (Chainlink, Pyth)Instant at expiryOracle network"Will BTC be above $X on date Y"
Optimistic oracle (UMA)Hours to daysBonded proposers, token-holder disputesGeneral events with public answers
Committee or multisigFastNamed reportersPrivate or niche markets, early-stage platforms
Platform as resolverFastTrust the operatorRegulated venues, where the operator is accountable
Reality.eth with arbitrationDaysEscalating bonds plus arbitratorLong-tail community markets

Hyperliquid also added on-chain outcome markets. Its HIP-4 went live on mainnet in May 2026 with fully collateralized binary contracts that settle to 0 or 1, starting with a small number of team-launched markets.

Writing markets that resolve cleanly

Most disputes come from bad market wording, not bad oracles. Rules we use:

  • Name one primary resolution source and a fallback, by URL.
  • State the exact time and time zone the outcome is measured.
  • Define edge cases in advance: postponed games, partial results, revised data, ties.
  • Avoid questions whose answer depends on interpretation ("Will X be considered...").
  • Keep a public log of clarifications, and never change rules after trading starts.

Wording is a product feature. Put an editor and a review step in the market creation workflow.

The components of a prediction market platform

ComponentWhat it does
Outcome token contractsSplit, merge and redeem positions against collateral (CTF or your own ERC-1155)
Exchange or market makerCLOB settlement contract, LMSR contract, or parimutuel pool
Matching engineOff-chain order matching for a CLOB, with signed orders and cancellations
Oracle adapterConnects each market to UMA, a price feed, or a committee; handles disputes and fallbacks
Market factory and adminCreates markets, sets fees and rules, pauses markets, handles edge cases
Market creation workflowDrafting, legal and wording review, resolution source checks, scheduling
IndexerPrices, volume, positions, PnL, market history
FrontendMarket pages, charts, order entry, portfolio, redemption
Wallet and onboardingEmbedded wallets, fiat on-ramp, gas sponsorship so users never see gas
Compliance layerGeoblocking, KYC where needed, sanctions screening, market-category restrictions
Market-making and liquidityAPIs for market makers, AMM subsidies, or incentive programs

Onboarding matters more here than in most DeFi products, because prediction market users are often not crypto-native. Embedded wallets and gas sponsorship, covered in our account abstraction guide, close most of that gap. For the indexing side, see The Graph vs a custom indexer.

How to build a prediction market, step by step

  1. Decide the legal structure first. Which users, which jurisdictions, which market categories. This decides whether you can launch at all and what you must build for compliance.
  2. Pick the mechanism. CLOB if you can get market makers, LMSR if you need many markets with guaranteed quotes, parimutuel for simple products.
  3. Pick the chain and collateral. Cheap, fast chains and a liquid stablecoin. See our stablecoin payments guide for collateral and on-ramp choices.
  4. Choose resolution per market type. Price feeds for price markets, an optimistic oracle or committee for events.
  5. Build contracts. Outcome tokens, exchange or AMM, oracle adapter, factory. Fork the CTF where it fits rather than rewriting it.
  6. Build matching, indexer and frontend. Signed-order matching for a CLOB, real-time prices, portfolio and redemption flows.
  7. Build the market operations tools. Drafting, review, scheduling and dispute handling. Teams underestimate this every time.
  8. Audit, then launch with a small set of markets. Ten well-written markets with liquidity beat five hundred empty ones.

Prediction markets are a fast-moving regulatory area, and the answer depends on where your users are.

United States. The CFTC regulates event contracts traded on designated contract markets. Kalshi has operated as a CFTC-regulated exchange since 2020. Polymarket bought QCEX, a CFTC-licensed exchange and clearinghouse, in July 2025 and launched a separate US app, which dropped its waitlist for iOS users in May 2026. The CFTC published an advance notice of proposed rulemaking in March 2026 and followed it in June 2026 with a proposed rule on public interest determinations for event contracts, which commentators read as presumptively permitting many sports contracts. States are pushing back: several sent cease-and-desist letters, and in April 2026 the CFTC and Justice Department sued Arizona, Connecticut and Illinois to defend federal jurisdiction, with more suits after. Court rulings are split and appeals are still running. In practice, offering event contracts to US users means either operating through a registered DCM or partnering with one. A permissionless dApp with a US-facing frontend is a serious enforcement risk.

European Union. On 3 July 2026, ESMA stated that event contracts which qualify as MiFID II financial instruments fall under the existing national bans on selling binary options to retail clients, and that the "prediction market" label doesn't change the classification, as PwC Legal summarized. Contracts on events outside MiFID's list of underlyings, such as elections or sports, tend to be treated as gambling under national law instead. France's gambling regulator ordered internet providers to block Polymarket in July 2026, and the platform is restricted in more than 30 countries. MiCA covers crypto-assets, so tokenized positions add another layer to check. A pan-EU, retail, crypto-native prediction market is very hard to launch legally today.

Elsewhere. Some jurisdictions license prediction markets as betting, some as derivatives, some not at all. Geoblocking, wallet screening and KYC are standard regardless.

How much does it cost to build a prediction market?

These are our planning ranges for build work, by scope. They exclude audits, licences, legal work, market-making and liquidity subsidies.

ScopeWhat it includesTimelineBuild budget (USD)
PoCParimutuel or LMSR markets on testnet, admin resolution, basic UI3–6 weeks$20k–$50k
MVPConditional tokens, LMSR or simple order book, UMA or price-feed resolution, embedded wallets, indexer, market admin8–14 weeks$80k–$200k
ProductionHybrid CLOB with off-chain matching, market-maker APIs, multiple oracle paths, dispute tooling, compliance layer, mobile-ready UI5–9 months$250k–$600k+

Other published estimates vary widely. RaftLabs puts an on-chain prediction market MVP with an order book, wallet and oracle resolution at $200,000 to $450,000 over 20 to 30 weeks. At the other end, "Polymarket clone scripts" sell for a few thousand dollars. A clone script gives you screens and contracts. It doesn't give you an audit you can rely on, a resolution process users trust, market makers, or a legal structure, and those are what decide whether the platform survives.

For the audit, Sherlock's 2026 reference puts a mid-complexity DeFi protocol at $40,000 to $100,000. A fork of the CTF plus a small exchange contract sits at the lower end. For a ballpark on your own scope, use our cost calculator, and see blockchain app development cost for other project types.

Frequently asked questions

How do prediction markets make money?

Mostly through trading fees on matched orders, fees on winnings, or the spread on an AMM. Some charge market creation fees or earn yield on collateral. On an order book, volume depends on market makers, so many platforms share fees or pay rebates to them.

Can I build a Polymarket clone?

Yes, technically. Polymarket's original CTF Exchange is open source under MIT and the Gnosis Conditional Tokens Framework is public, so the contracts are a known quantity. The hard parts are resolution you can defend, liquidity on day one, and a legal structure for your target users.

How are prediction markets resolved?

Price-based markets usually resolve from an oracle feed at a set time. Event markets on crypto platforms often use UMA's optimistic oracle, where a bonded proposer posts the result and anyone can dispute it to a token-holder vote. Regulated venues typically resolve markets themselves under published rules.

Are prediction markets legal?

It depends on the jurisdiction and the contract. In the US, event contracts are legal on CFTC-regulated exchanges, though several states are challenging sports contracts in court. In the EU, many event contracts fall under retail binary options bans or national gambling law, and Polymarket is blocked in several member states.

How long does it take to build a prediction market?

A PoC takes 3 to 6 weeks and an MVP with conditional tokens, oracle resolution and an indexer takes 8 to 14 weeks of build time, plus an audit. A production platform with a hybrid order book and compliance tooling takes 5 to 9 months.

Build your prediction market with us

We build dApps with embedded wallets, custom contracts, indexers and the admin tools that keep them running. If you're planning a prediction market, we can help you pick a mechanism and resolution model that fit your legal setup, and scope a realistic first version. See our decentralized application development services or contact us.

Keep reading

Ready when you are

Tell us what you are building and where you are stuck. You get a reply from an engineer and a scoped estimate for a PoC, MVP or full launch. We reply within 24 hours, usually the same business day.

23stud.ioWeb3 and blockchain development studio. Advice, development, team extension or complete project delivery.
Copyright © 2026 23stud.io. All rights reserved