Next Gen Prediction Markets w/ Ivo Crnkovic-Rubsamen

Ivo Crnkovic-Rubsamen is the founder of Pascal, a prediction market venue built on Solana that uses an off-chain matching engine while holding user funds in on-chain smart contracts. He previously worked at Bridgewater and D. Shaw before becoming one of dYdX's first hires with a real trading background and eventually its CEO. Pascal's off-chain matching engine is a deliberate choice to preserve traditional HFT sequencer guarantees and avoid MEV complications, a decision Ivo acknowledges is controversial given Solana's conventions. He is skeptical that fully on-chain matching on Solana can scale to genuine high performance.

Before building Pascal, the three-person founding team spent roughly three months trading on Kalshi and Polymarket to understand the venues from the inside. Their overall P&L was marginally positive. A strategy of maximizing liquidity reward extraction by posting orders at the touch across many markets failed badly, with a pennies-in-front-of-a-steamroller payout profile. The team concluded that successful small trading teams on these venues are modelers first and engineers second, and that their own lack of deep modeling expertise was a significant limitation over that window.

Polymarket had serious structural problems roughly a year before this recording. A nonce-based attack allowed traders to send a conflicting transaction with higher gas directly to the Polygon sequencer, invalidating a matched trade before it landed on-chain. Polymarket's matching engine sent fill notifications over websocket before trades were finalized, and Polygon takes approximately eight seconds to finalize with block times of around two seconds, creating an exploitation window. Polymarket has since hired strong people and upgraded its smart contracts, largely resolving the nonce attack issue though some tail cases remain. Polymarket's yes-no token architecture also requires buying 200 no tokens to move from a 100 long yes position to a 100 long no position, making liquidity provision far more capital intensive than necessary. Pascal eliminates this with a long-short model where selling 200 from a 100 long position simply results in a 100 short, though Ivo concedes this framing is slightly less intuitive to average users.

Polymarket's liquidity rewards system is explicitly modeled on the dYdX liquidity rewards formula. The optimal farming strategy is to avoid queue priority so orders are never actually filled while still collecting the reward, a behavior Ivo says was also observed on dYdX. Large sophisticated firms like SIG will not seriously market make on small novelty prediction markets, meaning large price swings in those markets almost certainly reflect insider information. Ivo views Polymarket's viral novelty markets as best evaluated as a customer acquisition cost versus lifetime value calculation and predicts they have no serious long-term future beyond being a growth hack. On Kalshi, queue priority is the most important edge factor because tick sizes are large and order flow is soft, making front-of-queue positions against retail flow highly profitable.

Pascal's core thesis is that prediction markets will evolve beyond sports gambling into serious financial instruments used by traders and real-economy counterparties managing genuine risk. Ivo says Pascal has been asked to quote every block trade that has ever printed on Kalshi, yet in his estimation Kalshi has printed fewer than ten real risk-exchange block trades. He argues that variance in profit and loss only matters to a leveraged company, either through debt or operating leverage, and that unleveraged companies have no incentive to pay expected value for risk reduction. He cites a barge company on the Mississippi whose year-over-year profitability is driven primarily by days of sufficient water level, and argues that liquid rainfall markets on prediction platforms provide a pricing reference enabling more bespoke risk transfer for counterparties like that company. He also describes large data centers increasingly preferring derivatives contracts tied to water level thresholds over traditional flood insurance for GPU assets, because the derivatives model pays out instantly rather than requiring a lengthy claims assessment, though he notes he has heard about such deals rather than seen direct deal flow.

On market structure, Pascal charges fees three to five times lower than Polymarket depending on category and pays half of the taker fee to the maker atomically on every fill, whereas Polymarket aggregates and pays maker rebates on a delayed and opaque schedule. Ivo argues that exchanges only make sense with a heterogeneous user base having different utility functions and time horizons. If users are modeled purely as gamblers indifferent to price, a sportsbook is a better business model because the operator can internalize all flow. He characterizes Kalshi's success as largely regulatory arbitrage through CFTC regulation and says that if the regulatory environment does not improve, sportsbooks will eventually outcompete venues like Kalshi, which he believes needs to offer perps on equities or similar instruments to compete long term.

Ivo identifies perps and event contracts as the two asset classes that will continue to grow, with perps best suited for assets with a liquid and readily observable oracle and event contracts better suited when no oracle exists. Event contracts cannot result in liquidation but are fully collateralized, making capital efficiency a major challenge. Pascal is building a prime brokerage and leverage layer separate from its core matching engine to address this. He contrasts this with crypto perp exchanges like Bitmex, which had to build leverage provision into the exchange itself because no external prime brokerage layer existed, leading to mechanisms like auto-deleveraging that would never occur in traditional finance.