Prediction markets: from curated platforms to open protocols

Author: Noveleader; Source: Castle Labs; Compilation: Shaw, Golden Finance

Over the past 18 months, monthly trading volume in prediction markets has surged from approximately $2 billion to over $30 billion, rapidly evolving from a niche track to a mature industry ecosystem.

Prediction markets are gradually integrating multiple attributes: they are both information trading markets and specialized hedging tools, while supporting users in trading and predicting outcomes for various events such as sports, politics, macroeconomics, and more.

The range of trading targets is extremely broad, spanning from niche topics like "what a certain podcast will say" to major events like "the Fed's interest rate decision." Users can essentially trade the outcome of any event.

The current industry growth has relied almost entirely on a platform-controlled model: platforms designate a few entities to decide which trading markets to list. Leading platforms like Polymarket and Kalshi both adopt this model, strictly controlling the tradable targets, and the market has responded with positive trading volume feedback.

While the platform-controlled model is workable, it deviates from the original design intent of early products in this track. The original goal of prediction markets was to enable permissionless market creation—anyone could create a trading market on any topic. However, most products built on this concept (Augur, Omen, Zeitgeist, etc.) have failed, continuously facing multiple challenges such as insufficient liquidity, event outcome resolution, creator incentives, and regulatory compliance.

The industry thus concluded that a fully open permissionless model cannot scale, but this assertion is now being re-examined. A new batch of products are entering the scene, attempting to solve the various pain points left by earlier products by optimizing oracle infrastructure, building scalable liquidity mechanisms, and refining product design.

This report reviews the evolution of prediction markets, analyzes the reasons for the initial failure of the permissionless model, the reasons for the success of the platform-controlled model, and how current products are reverting to the permissionless nature of prediction markets. It also uses Limitless as a case study: the platform recently opened user-generated market (UGM) functionality, and this article will detail the product adjustments made to accommodate user-created markets.

Early Prediction Markets: The Permissionless Development Path

During the 2024 U.S. presidential election, Polymarket and Kalshi provided event odds with far greater accuracy than other channels, bringing the value of prediction markets to widespread attention for the first time. Since then, the business boundaries of prediction markets have expanded significantly, no longer limited to election predictions.

Since the beginning of this year, the nominal trading volume of prediction markets has consistently exceeded $20 billion, surpassing $40 billion last month; industry trading volume is mainly contributed by the two leading platforms, Polymarket and Kalshi, forming a duopoly in the industry.

Prediction markets were among the earliest experimental foundational components in the crypto space, relying on the core concepts of decentralization and permissionless participation. Early products all adopted a permissionless construction model.

Augur was one of the first products in this track to launch, officially going live in July 2018. The project team was founded in 2014, and its initial coin offering (ICO) was the first in the track, raising $5.5 million.

On the Augur platform, anyone could create a prediction market without permission. This permissionless architecture quickly exposed numerous risks: within weeks of launch, the platform's frontend featured markets involving political assassinations, plane crashes, and other legal red lines. In addition, various technical defects followed, such as high Ethereum Gas fees, settlement adjudication delays (in case of disputes, the adjudication process could take up to 90 days).

A distinctive feature of the product was its native token REP, where holders could stake tokens to report the final outcome of events.

In 2026, the Lituus Foundation rebooted the Augur project, transforming it into an event outcome resolution infrastructure that could be called upon by other prediction markets, shifting the competitive focus to underlying infrastructure. The new resolution engine mechanism: participants stake their own funds to bet on a certain outcome; whenever a dispute occurs, the capital threshold required for each party's staking becomes higher, increasing the cost of deliberately lying or maintaining false conclusions.

Gnosis was another early participant in the track, launching a conditional token framework in 2017, allowing anyone to create markets and convert event outcomes into tradable tokens. Polymarket later adopted this framework. Gnosis eventually shut down its own prediction market product, facing challenges highly similar to Augur: high Ethereum Gas fees, insufficient scalability, and primitive user tools.

In 2020, Omen launched based on the Gnosis framework, supporting permissionless market creation and equipped with an automated market maker (AMM) mechanism. However, its core pain point was fragmented liquidity: anyone could create markets on any niche topic, leading to hundreds of highly similar trading pools for the same event, with the vast majority of pools having no liquidity at all.

In addition, the platform faced oracle-related issues. Omen used the decentralized external oracle Kleros for outcome resolution: Kleros uses crowdsourced jury voting, incentivizing jurors to follow the majority opinion, which may not necessarily equal objective facts or the true outcome of events. At the same time, the resolution process was slow, and Gas costs remained high.

These were just technical and liquidity issues; platform economic models and product design flaws were also significant.

In early 2018, Stox launched a permissioned prediction market covering sports, finance, and news events, raising $33 million in its ICO. The core reasons for Stox's failure were the lack of a profit model and a mismatched token economic mechanism. Although the platform charged transaction fees, the revenue was insufficient to support market maker incentives; the platform's native token STX holders were supposed to share platform revenue through fee distribution, but there was no on-chain mandatory distribution mechanism, failing to sustainably attract users and liquidity. Technically, Stox's resolution oracle was highly centralized and entirely dependent on the operating company, casting doubt on decentralization.

Another permissionless platform, Hedgehog Markets, was built on the Solana blockchain, allowing users to create markets independently. The project launched "principal-protected prediction markets," where users deposit 100 or 1,000 USDC principal in exchange for game tokens used for trading predictions; the interest generated from the deposited USDC pool is distributed entirely to winning traders. While this model seemed innovative, it severely limited user profit potential—users could only earn interest returns, with principal not at risk. For investors willing to take principal risk and trade deeply, the mechanism had clear profit imbalance, significantly reducing willingness to participate.

All the above products exposed industry pain points from different dimensions: liquidity depletion, disorderly market creation, failure of event outcome resolution mechanisms, and a concentration of compliance risks.

Market Discovery Challenges

After the permissionless creation mechanism was opened, users could create an unlimited number of new markets, leading to numerous duplicate trading pools for the same event. Under major events, hundreds of markets with low liquidity, confusing descriptions, and overlapping functions often appeared.

The proliferation of redundant markets severely damaged the user experience, making it easy for ordinary users to confuse targets. Experienced traders would concentrate capital flows on the best-liquidity pools, but this problem could not be completely solved. The options were either to restrict the creation of similar markets or to optimize the frontend by hiding low-liquidity markets and only supporting manual user search.

Event Outcome Disputes

"The value of a prediction market depends entirely on its ability to resolve outcomes that reflect objective facts."

Augur's REP staking oracle and Omen's Kleros mechanism both incentivized stakers to vote with the majority. Similarly, Polymarket's use of the UMA optimistic oracle had the same resolution flaw: UMA token holders vote to finalize the outcome, but these voters themselves can trade on Polymarket, making them prone to bias toward a certain outcome to benefit their positions, creating inherent interest bias in voting.

Taking Polymarket as an example, resolution power is currently highly concentrated among nine whale addresses, whose voting results consistently align with the eventual winning outcome. These large holders can easily manipulate market trends to profit from their own trades.

Recent typical case: the MicroStrategy Bitcoin sell-off prediction market. From May 26 to 31, 2026, the company did sell 32 Bitcoins, and the market should have been resolved as "Yes"; even after two rounds of dispute appeals, the final resolution was still "No."

Liquidity Fragmentation

The permissionless market creation model causes liquidity fragmentation: users create multiple competing markets around the same theme, splitting overall funds apart. The biggest challenge Omen faced was exactly liquidity fragmentation, with many users creating highly similar market targets.

There are two approaches to solve this problem: one is to tighten market creation review mechanisms, e.g., restrict to only one market per event; the other is to handle it through frontend optimization, aggregating all related markets under the same theme on one page, with the platform prioritizing the best-liquidity pool, allowing users to choose their trading target.

Manipulation Risk

Prediction markets themselves are highly susceptible to manipulation through capital or information means. Users can significantly influence market prices through large positions, while simultaneously holding significant voting power in oracles to guide resolution results in favor of their positions.

Both permissioned and permissionless models have this vulnerability, but in permissionless models, users can create unlimited new markets, drastically expanding the attack surface.

Recently, a Google employee was investigated by the U.S. Commodity Futures Trading Commission (SEC) for trading Google search-related prediction targets on Polymarket. The employee used insider information to arbitrage, knowing the final outcome beforehand, profiting millions of dollars.

Regulatory Compliance Issues

The U.S. Commodity Futures Trading Commission (CFTC) classifies event-based contracts as futures contracts. According to the Commodity Exchange Act, such products must be registered on a designated contract market (DCM).

Most early prediction markets did not meet compliance requirements, which became a core obstacle to industry scale. In 2022, Polymarket was fined $1.4 million for non-compliant operations. The U.S. market has strong demand for sports betting and event trading, with a huge market size. Therefore, Kalshi has always adhered to a compliance-first approach, while Polymarket also acquired QCEX in 2025 to complete U.S. regulatory registration.

Lack of Market Creator Incentives

Early permissionless market creation products did not design supporting economic incentives, failing to consider that market creators could derive revenue from operating targets. This led to an imbalance of interests between the platform and market creators: even if a market's popularity and liquidity increased, the creator lacked motivation and received no economic reward.

Today's leading legacy platforms have recognized these pain points and shifted to a permissioned market creation model to resolve related issues.

The Permissioned Market Review Model: The Development Path of Polymarket and Kalshi

Polymarket launched in 2020, and Kalshi launched in 2021. The two platforms initially took different development paths, but recently their strategies have converged in terms of compliance layout and competition for the U.S. market.

As the industry developed, Kalshi and Polymarket formed a fierce competitive relationship.

Before mid-2025, Polymarket consistently held an 80% market share. Since then, Kalshi has gained a competitive advantage through partnerships with platforms like Robinhood, now accounting for the majority of prediction market trading volume.

In terms of fee revenue, Kalshi's current annualized fee revenue is about $2 billion, while Polymarket's is only about $300 million. Additionally, in recent funding rounds, Kalshi was valued at $22 billion (current target valuation $40 billion), while Polymarket was valued at $15 billion; from a price-to-sales (P/S) perspective, Polymarket appears overvalued compared to its competitor.

The two platforms have reached their current scale because they have largely solved the various pain points of early prediction markets mentioned earlier.

First, in terms of compliance, Kalshi's compliance-first strategy has been significantly effective, helping it achieve rapid growth in the U.S. market. Polymarket also launched the U.S. compliant version Polymarket U.S., gradually rolling out operations in phases.

In terms of market target management, both platforms have dedicated market review teams, adopting a permissioned management model, effectively avoiding issues like liquidity fragmentation and duplicate markets.

Although permissioned platforms have achieved success, this does not mean that the permissionless model cannot coexist in the track. A large number of users still have the need to create markets independently: they want to set up niche topic markets of interest, share in market creation fee revenue, and participate in providing liquidity.

Early permissionless prediction markets exposed many defects, and leading platforms have resolved these difficulties through permissioned models.

Even though leading platforms have operated for years and expanded significantly, users still cannot independently create and trade markets on any topic. But the ultimate vision of prediction markets is to allow anyone to use capital bets to create objective and credible price references for various events.

Against this backdrop, as industry solutions continue to improve, the permissionless market creation model is experiencing a strong resurgence.

The next section will analyze the current development status of the permissionless self-built market track and the feasible path for user-generated markets (UGM) to scale.

The Shift to Permissionless: A New Generation of Prediction Markets

Over the past year, the permissionless market creation track has continued to develop, with a large number of new projects entering, including Melee, HIP-4, XO Market, etc.; the established platform Limitless has also expanded its business by launching user-generated markets (UGM), entering this track.

Limitless recently launched permissionless market functionality. The platform has not yet fully opened permissionless creation permissions; currently, it only allows the creation of crypto-related theme markets, with plans to gradually expand to other categories, thereby gauging market demand while simultaneously scaling the business.

In addition, market creators on the platform can receive 50% of the fee revenue from their markets, deeply aligning the interests of the platform and market creators.

HIP-4, i.e., Hyperliquid restricted markets, sets a capital entry threshold: market deployers must stake 1 million HYPE tokens to qualify for market deployment slots. Developers can set a maximum of 50% split fee rate on top of the Hyperliquid base fee (this feature was just launched in early May and currently waives all fees to stimulate trading). The high deployment threshold fundamentally eliminates junk and redundant markets.

HIP-4 markets also have an architectural advantage: they run natively on Hyperliquid's underlying Hypercore, sharing the same order book, account system, and margin engine as the platform's spot and perpetual contracts. Traders do not need to split funds across multiple accounts to participate in prediction markets and hedge their portfolios.

Both types of platforms use an order book trading model, with liquidity initially provided by market makers (MM). Other platforms adopt differentiated liquidity solutions.

XO Market uses a liquidity-sensitive logarithmic market scoring rule (LS-LMSR) automated market maker model, which is an upgraded version of the standard LMSR model commonly used in most prediction markets.

The core difference lies in liquidity management: Under the standard LMSR model, market creators must pre-set fixed liquidity parameters, effectively estimating the trading volume the market can attract in advance. Setting the parameter too low causes prices to be overly sensitive to single trades; setting it too high requires significant capital. LS-LMSR makes the liquidity parameter dynamically adjustable, with market depth automatically adapting to trading activity, eliminating the need for manual presets.

When creating a market, users must also provide initial startup liquidity themselves. This measure can curb junk markets and alleviate liquidity depletion issues, hence the platform is also called a "belief market"—participants must invest capital to express their judgment.

In the event outcome resolution phase, XO uses a unique three-tier resolution mechanism. The first tier is an AI priority channel, relying on MODRA (Market Outcome and Dispute Resolution Agent) to automatically and quickly determine simple events with clear facts; if disputes arise, it enters the second tier Senate jury for manual review, and if unsatisfied, an appeal can be made to the Supreme Court. The platform has accumulated a total trading volume of over $250 million, listed more than 2,800 trading markets, and completed over 30k trades.

Another prediction market project, Melee, iterates on the pari-mutuel betting model, naming its solution PMM. Traditional pari-mutuel betting pools gather all betting funds, with final payout determined by total bets and the number of winning bets. This model is common in scenarios like horse racing and has a hard time limit: traders cannot enter or withdraw funds after the event starts.

The pari-mutuel betting model has many inherent flaws: users can only redeem funds after the market settles, cannot exit mid-way, leading to insufficient trading flexibility and repelling many traders. Even if the market is still open, traders wanting to exit positions can only buy opposite outcome tokens; however, prediction market prices fluctuate in real-time, and position values change constantly. Additionally, this model requires markets to close before the event starts—during the event, participants may concentrate bets on the already advantageous side, exploiting the mechanism's flaws for arbitrage.

Melee is optimizing the pari-mutuel betting model by supporting continuous trading to eliminate the above shortcomings, but it has not yet officially launched.

Products like Xmarket use a unique initial liquidity mechanism: the minimum startup capital for creating a market is only $1, but the market will only officially launch after reaching a soft liquidity threshold of $100; if the threshold is never met, all participant funds are returned. This mechanism filters meaningless junk markets at the fundamental level, identifying targets with genuine attention and gauging real user demand. However, the threshold value is low and can easily be manually manipulated to meet it; even so, the mechanism can effectively activate initial liquidity and market trading activity.

The chart below will visually compare the solutions of the above products for the various pain points exposed by the first-generation prediction markets. The industry has evolved multiple technical routes for the five core challenges of market creation, outcome resolution, liquidity provision, economic incentives, and compliance paths, fully demonstrating the positive innovation and diverse experimentation atmosphere among track practitioners.

User-generated markets (UGM) have encountered many difficulties in the past, and there is currently no conclusive evidence that any single solution works best—so far, no user-generated market has achieved large-scale adoption.

The next section will deeply analyze a recently launched product, Limitless, focusing on its UGM operating model and how it addresses the various industry pain points identified earlier.

Financial Prediction Markets: How Limitless User-Generated Markets Operate

Limitless entered the prediction market track in 2024, and its business scale has grown steadily. Its development dividend also benefits from the platform using its native token as a vehicle for user incentives.

Limitless previously used a permissioned market creation process, with all trading targets designed internally by the platform team. The platform has a standardized outcome resolution mechanism: financial markets such as cryptocurrencies, stocks, commodities, and forex are settled using data from Pyth and Chainlink oracles; sports, politics, and other categories are manually resolved by the Limitless team. If the final outcome cannot be determined, user funds are fully refunded.

Leveraging its already well-built underlying infrastructure, the platform is expanding its business by launching permissionless market creation functionality.

On June 2, 2026, Limitless launched its first permissionless market category. To proceed cautiously with scaling, the platform did not fully open market creation for any topic but instead adopted a standardized template and a strategy limited to financial categories. In the initial phase, market creators could only choose targets from a designated pool of crypto assets (Bitcoin, Ethereum, Solana, XRP, Dogecoin), set interval-limit price targets (fluctuation range -5% to +5%), and select a trading duration from 15 minutes to 1 day. The core purpose of this standardized template mechanism is to eliminate redundant markets with vague descriptions and odd wording.

In the first month after launch, the cumulative trading volume of such permissionless markets reached $2.2 million.

User-generated markets (UGM) run concurrently with officially created markets, making Limitless a hybrid prediction market platform. Currently, only crypto category UGM are open; after verifying market demand and stabilizing platform operating standards, the team will gradually introduce other categories.

As the scale of such UGM grows, they can evolve into customized hedging tools, allowing users to create dedicated trading markets based on their perpetual/spot positions. This application scenario will further expand after the platform adds more assets and prediction categories such as stocks, commodities, sports, and e-sports.

Early prediction markets exposed various typical pain points, and Limitless uses a differentiated set of solutions to address them, explained by dimension as follows:

• Market Creation: Limitless adopts a gradual strategy, initially opening only a few financial targets, primarily the five crypto assets: Bitcoin, Ethereum, Solana, XRP, and Dogecoin. Users set price intervals and trading durations through standardized templates, significantly reducing obscure and odd markets, and avoiding disputes in the later settlement phase.

• Event Outcome Resolution: Since the initial phase only includes crypto-related markets, the platform reuses the same oracle system as the permissioned markets, connecting to data sources like Chainlink and Pyth, ensuring consistent and stable price feeds across the platform. Fully automated oracles do not require continuous manual monitoring, making market creation and settlement processes smoother and more flexible.

• Liquidity Provision: Limitless uniformly injects initial startup liquidity for user-generated markets. All UGMs are built on a central limit order book (CLOB) architecture, with the underlying infrastructure fully open, allowing any market maker to enter and provide liquidity, achieving efficient price discovery.

• Market Creator Incentives: The success of user-generated markets hinges on the creator's economic reward model—i.e., how creators profit. Without substantial revenue sharing, creators will not build markets properly and will instead create many targets with vague information, harming market search experience and failing to attract trading volume. The alignment of interests between the platform and creators is key to long-term stable operation. On Limitless, creating a market requires paying 100 to 1,000 LMTS platform tokens (the fee increases with longer trading durations); creators receive 50% of the fees generated by the market. The team is also collecting user feedback to explore more flexible pricing models.

• Market Search: Currently, user-generated markets are limited to the price movements of a few crypto assets, with clear restrictions on asset type, trading duration, and price movement range, reducing market fragmentation and highly similar duplicate targets from the source.

• Regulatory Compliance: In early May this year, Limitless submitted an application to the U.S. Commodity Futures Trading Commission (CFTC) to become a federally regulated derivatives exchange (the CFTC has deemed the application complete and entered formal review). Once approved, the platform will have a clear compliance path, enabling large-scale expansion in the U.S. prediction market track, directly competing with Kalshi, compliant Polymarket, and Crypto.com derivatives business.

Limitless's mechanism provides comprehensive solutions to the various defects of previous-generation prediction markets: standardized templates ensure platform order, eliminating low-quality markets with vague information; the creation fee mechanism filters out junk and spam markets from the root; the 50% fee sharing gives creators real motivation to actively operate and stimulate trading; meanwhile, the platform is awaiting CFTC approval, having a clear compliance route, offering differentiated advantages over competitors.

However, the creation fee also introduces the classic "chicken-and-egg" dilemma: users must pay to create a market, so they must weigh the upfront cost against the potential revenue from the 50% fee split. Revenue depends on sufficient trading volume, while the creation cost is fixed.

Under the current Limitless model, the fee for creating a market is fixed and only denominated in the platform's native token, LMTS. While this mechanism effectively intercepts junk markets, the platform could consider introducing a dynamic fee mechanism: lowering fees when demand for certain asset or market creation is low, and raising them when demand is high, optimizing supply-demand balance.

Future Development Outlook

The prediction market track started early, with first-generation products like Augur and Gnosis emerging during the crypto ICO boom. These early projects faced highly similar industry pain points, concentrated in four areas: liquidity fragmentation, slow settlement and frequent disputes, lack of market creator incentives, and prominent compliance risks.

Now, the new generation of projects in the permissionless market creation track are clearly aware of these shortcomings and adopt differentiated product designs to specifically address the various difficulties existing in current prediction markets.

XO Market requires creators to inject initial liquidity themselves, paired with a multi-tier resolution mechanism aiming for the most objective and accurate event outcomes; Melee reconstructs the traditional pari-mutuel betting model to achieve continuous trading; Xmarket sets a soft capital threshold to filter out meaningless junk markets; HIP-4 sets a high token staking entry threshold, only allowing quality markets to launch; Limitless introduces a 50% fee sharing mechanism and charges market creation fees, perfecting the creator revenue system while preventing the proliferation of junk targets.

However, the real core difference between these projects lies not in their individual solutions but in the completeness of the entire solution across different dimensions. A product that only solves the liquidity problem but fails to handle resolution properly, or that only optimizes the settlement mechanism but lacks a creator incentive system, will struggle to achieve scale.

Based on the above background, the permissionless user-generated market track is facing a new development opportunity worthy of continued attention.

In the long run, the market will screen out the best design solution suitable for large-scale implementation, creating the optimal trading platform that can cover various niche sub-tracks and restore objective facts—which is exactly the core original intent of the underlying architecture design of prediction markets.