Prediction markets had a very good few years. Then 2026 happened, and the problems that were always baked into the architecture started becoming impossible to ignore. A US Army soldier made $400,000 on Polymarket using classified intelligence about a military operation. Politicians bet on their own races on Kalshi. A Vanderbilt University study examined 2,500 markets and concluded that Polymarket, the largest platform in the world, encourages herding behaviour rather than genuine information aggregation. None of this is a scandal about bad actors. It is a diagnosis of a design that was always going to produce these outcomes. And understanding why helps you understand exactly what Bench is building instead.
Start With What Prediction Markets Actually Are
Prediction markets are probability machines. You pick a binary question: will this candidate win, will this company hit its earnings target, will this team make the finals. Participants buy and sell contracts. The market price reflects collective probability. In theory, prices move when new information enters, and the aggregate of many individual bets produces a more accurate forecast than any single analyst.
The theory is not wrong. For events with a lot of publicly available information, markets can aggregate distributed knowledge effectively. The 2024 election markets were genuinely more accurate than most polling averages. There is real signal in a large group of people putting real money behind their beliefs.
But prediction markets are built for a specific kind of question. They need the outcome to be binary. They need the resolution to be objective and observable. And critically, they need the information driving the market to be distributed across many participants rather than concentrated in a small group of genuine experts.
When those conditions are not met, the model starts to fail. And a very large category of the most valuable decisions in the world fails all three of those conditions simultaneously.
The Herding Problem Nobody Wanted to Talk About
The Vanderbilt researchers found something specific that gets to the core of the structural issue. When stakes are public and visible, participants do not primarily price based on their own genuine assessment. They price based on what they observe other participants doing. This is herding. And herding produces a reflection of existing consensus, not new information.
Think about what that means in practice. If you are a genuine expert with a contrarian view, putting your stake publicly into a prediction market is a bad strategy. You reveal your conviction to the whole market before it pays off. Sophisticated participants with real information either do not enter at all, wait until the last possible moment to enter, or enter in a way that obscures their actual position.
The result is that the participants with the most genuine expertise systematically underparticipate in prediction markets. The people you most want to hear from are the least incentivised to speak clearly.
“The participants with the most genuine expertise are systematically the least incentivised to speak clearly in a public market. Bench inverts this dynamic entirely.”
A Different Question Entirely
Here is the distinction that matters. Prediction markets ask: what will happen? That is a forecasting question. Opportunity markets ask: what should I do? That is a decision question. They sound similar. They require completely different architectures.
A music label trying to decide which artists to sign does not want a probability. Will this artist go platinum? Maybe. Probably not. Most artists do not. The useful question is: among the fifty artists we are considering, which ones do our best scouts, the people with genuine A&R instincts and market knowledge, actually have conviction about? That is not a prediction. That is a judgment call from someone who knows something.
A startup trying to hire a senior engineer does not want the market consensus on whether a candidate will succeed. It wants to know which candidates have genuine advocates inside the professional community, people who have worked alongside them or observed them closely, who are willing to put real stake behind their conviction in a way that signals genuine belief rather than politeness.
An investment firm sourcing deals in a niche sector wants the entrepreneurs and operators who are actually inside that sector to surface the most promising opportunities, not the crowd that is guessing from the outside.
None of these use cases work with public stakes. The moment a music industry scout publicly stakes on an emerging artist, they have just told every competing label what they know. The signal evaporates the moment it is shared. The expert has no incentive to participate honestly in a public market because honesty is expensive.
How Bench Works
Bench solves this with a structure that is worth understanding mechanically, because each piece of it addresses a specific failure mode from the prediction market model.
A sponsor, a team, a label, a founder, a VC, sets up a market with a prize pool and an initial set of options. They define what they are looking for. This is the decision they need help making.
Participants with relevant expertise stake tokens on the option they genuinely believe is the right answer. Critically, scouts can also introduce entirely new options that the market creator never considered. The more you stake, the larger your share of the prize pool if your option wins.
Every stake is encrypted via Arcium’s MPC network from the moment it is submitted. Nobody sees who staked, on what, or how much until the window closes. No herding. No front-running. No signal leak. The encryption is cryptographic, not a policy rule.
After the window closes, the market creator sees the aggregated results and selects the winning option. Everyone who staked on that option shares the prize pool proportionally, with a multiplier rewarding those who staked early. The creator gets private, high-conviction signal. The scout gets paid for their expertise.
Where This Actually Matters
The RTG asks for real examples. Here are five, each chosen because they represent a decision category where prediction markets structurally cannot work but Bench can.
Artist signing at a record label. A label has a shortlist of emerging artists and wants to know which ones their best industry scouts have genuine conviction about. A public prediction market here is useless. The moment a well-respected A&R figure publicly bets on an artist, competing labels see it and bidding wars drive up acquisition costs before the original label can close a deal. Bench lets scouts express conviction privately. The label gets the aggregated signal. Nobody outside the market sees who backed whom until the label has already decided.
Transfer market scouting. A football club wants to identify undervalued players before their competitors do. Every scout and agent in the industry has private assessments. Nobody shares them publicly because doing so drives up transfer fees. Bench creates a private channel for scouts to stake their conviction on which players are worth pursuing. The club gets an aggregated ranking based on real insider conviction. The scouts get compensated if the club acts on their recommendation and it proves right.
Senior technical hiring. A startup needs to fill a critical engineering role. Former colleagues, conference co-speakers, open source collaborators, all have genuine assessments of the candidates. None of them will share publicly because it creates social awkwardness and professional risk. Bench gives them a private mechanism to stake their conviction. The hiring team sees which candidates have the strongest advocacy from people who actually know the work. References become a market signal rather than a courtesy call.
Early stage deal sourcing. A fund wants to find the most promising companies in a niche sector before they become obvious. Domain operators who work inside that sector, founders who have exited adjacent businesses, technical specialists who can assess the underlying technology, all of them have genuine private assessments. Bench creates a market where they can stake their conviction confidentially, the fund sees aggregated conviction scores across opportunities, and the people who backed the right companies early get compensated for their insight.
Feature prioritisation for a software product. A founder wants to know which feature their most sophisticated power users actually believe matters most, without the bias that comes from asking publicly. In a public forum, people tend to endorse whatever seems to have the most momentum or whatever the founder visibly prefers. Bench gives power users a private stake mechanism. The founder sees what people are willing to put real conviction behind rather than what they are willing to say out loud.
Why the Encryption Is Not a Feature — It Is the Category
This is the part that I think most explanations of Bench underemphasise. Privacy in a Bench market is not a preference setting. It is what makes the category exist at all.
Without Arcium’s MPC layer, you do not have a private opportunity market. You have a public prediction market with a different interface. The herding problem comes back. The signal leak comes back. The incentive to share genuine expertise disappears. The category collapses back into everything that prediction markets already do badly.
Arcium’s MPC network does something specific: it runs computations on encrypted data. The nodes that process the stakes never see individual positions. The market results emerge from the computation without any single node, participant, or outside observer learning what went into it. The privacy is not achieved by trusting the platform. It is achieved by the mathematics of multiparty computation. This is the same technology that underlies Arcium’s broader encrypted execution infrastructure on Solana.
Bench co-founder Erik Plaumann described the gap this way: “There’s a significant gap between the people who hold valuable information and the decision-makers who need it. Most existing channels either leak that information or filter it through the wrong incentives, but encrypted staking on Arcium lets us close that gap. Users share knowledge and convictions, sponsors get credible signals, and no one can game the market by watching what others are doing in real time.” That is the category in one paragraph.
The early staker multiplier also matters more than it initially appears. Prediction markets reward people who wait, because the optimal strategy is to see how other stakes develop and then enter late when the outcome looks clearest. This produces consensus rather than genuine signal. Bench rewards people who stake early, because early conviction is the signal the market creator actually needs. Waiting until the answer is obvious helps nobody. The multiplier shifts the incentive structure toward genuine expertise shared at the moment it is most valuable.
A Category That Could Not Have Existed Before
The insider trading scandals on Polymarket and Kalshi in 2026 were treated as bugs to be fixed by policy. Kalshi banned politicians from betting on their own races. Polymarket revised its rulebook to prohibit trading on information obtained through breach of trust. These are the right responses to the legal exposure. But they are treating the symptom.
The underlying insight is that people with genuinely private, valuable information have always had an asymmetric advantage in public markets. The prediction market industry’s answer is: ban the insiders. Bench’s answer is: create a market where the insider’s advantage flows to the decision-maker who needed it, rather than to the insider trading against the crowd.
That reframe is not incremental. It is a different theory of what an information market is for. Prediction markets are for the public. Opportunity markets are for the person who needs to make a specific decision and is willing to pay for genuine expertise. The privacy layer is what makes it possible to route that expertise to the right destination without it leaking everywhere else.
Within its first week on Devnet, Bench attracted more than 4,000 signups. The first live markets are Hiring Markets and Investment Markets, which are exactly the two decision categories where the expert signal problem is most acute and where existing tools are most inadequate. That is not an accident. It is the product of building toward the problem rather than toward the analogy.
Prediction markets tell you what the crowd thinks will happen. Opportunity markets tell a specific decision-maker what their best potential advisors actually believe they should do. They are not the same category with different mechanics. They are built for different questions, different users, and different definitions of what useful information looks like.
That distinction is the whole product.