Introducing sdk.markets

April 23, 2026

Today, we're excited to share an early preview of sdk.markets. It's a toolkit for creating custom, parimutuel prediction markets on anything, built on @base and protected by @privy_io

While building Turf 2.0, we ran into a limitation that kept coming up: the markets available on existing prediction market apps don't cover the full spectrum of what sports fans actually care about.

The sports arguments happening in group chats. The hot takes that only make sense if you're in the group. The IYKYK moments that are endlessly debated but never priced or tracked.

That gap is what led us to build sdk.markets.

sdk.markets is an extension of what prediction markets already do great, built for dynamic, peer to peer, ephemeral micro-predictions.

No house, no spread, just your own markets, for anything. You create a question, people take sides, and the winning pool splits the losing pool. Pure conviction versus conviction, on whatever your community actually wants to bet on. The infrastructure handles the rest.

Why Parimutuel Markets?

Order book markets work well when you have deep liquidity and participants who are actively trading in and out of positions. The price discovery is clean, the odds are readable, and sophisticated participants can hedge and exit.

For the kinds of markets we've been building for Turf, that model broke down quickly. Most communities don't have the volume to support a functioning order book, and most questions don't need one. Parimutuel is simpler, fairer for thin markets, and doesn't require a counterparty to take the other side of your position. The pool is the counterparty. The tradeoff is that parimutuel has known weaknesses.

The "wait and see" problem

If pool composition is visible in real time, sophisticated participants will wait until the last possible moment to place large bets so they don't move the displayed odds and tip off the crowd. Show live pool composition and you inadvertently create an incentive to hide volume until close, which undermines price discovery entirely.

Our answer is shortened contract windows. Last season with Turf, we ran live prediction contests during NFL games and learned something unexpected: when everyone answers inside a 15 to 30 second window, the social dynamic changes entirely.

There's no time to watch pool composition and snipe. No time to wait for others to commit first. Everyone is betting on raw conviction simultaneously. It ends up being closer to a sealed bid auction than a continuous market, and it turns out to be manipulation-resistant by design. We stumbled into it through UX constraints and it works better than more elaborate solutions.

For longer-horizon markets we're exploring a few approaches. Snapshot-based locking fixes the share price per dollar at a defined cutoff, so early participants get better terms than late participants while the market stays open.

This follows the Dynamic Parimutuel Market (DPM) model where each dollar buys a variable share of the payoff and share prices rise as the market evolves, protecting early participants from last-second pile-ons. Delayed display shows pool composition with a lag, preserving some price signal without exposing real-time volume. Graduated late penalties reduce the share of the pool that last-second bets are entitled to, making sniping less worth it without hiding information entirely.

Each trades off transparency against manipulation resistance differently and the right answer depends on the market.

Resolutions with AI Oracles

The harder problem in any peer-to-peer market is who decides the outcome and how participants trust that decision. There are three resolution modes we built that can be run independently or combined:

1. Single admin. One trusted source resolves the market. A league manager, an event organizer, whoever the group designates. Simple, fast, works well for closed communities where trust is already established. The assigned admin determines the outcome but never has direct access to the funds.
2. Multi-admin consensus. Resolution requires agreement across multiple admins. You configure the threshold. Two of three agreeing is enough to resolve, or you can require unanimity. Useful for higher-stakes markets where you want redundancy and no single point of failure.
3. AI oracle. You point the market at a source and the oracle handles the rest. A Sleeper API endpoint that auto-resolves your fantasy league payouts. An ESPN page for a game result. An X account. A news URL for a world event. A podcast transcript. An on-chain data feed for a token price. Any URL can contain a verifiable outcome.

And you don't even need to know the right source upfront. Describe the market and the oracle suggests the most relevant sources automatically. Give it one source and it finds corroborating ones on its own. The oracle researches the outcome across all of them and resolves automatically. No one has to do anything after market creation.

Most prediction market infrastructure assumes a human in the loop for resolution. Removing that dependency is what makes markets on truly arbitrary questions tractable at scale. Describe a verifiable outcome and point to a source and a market now exists.

The SDK

Create a market on anything. Pick your resolution mode. Set a creator fee. Deploy.

We've been testing sdk.markets internally the past few weeks. It's powering some of the experiences we're most excited about for Turf 2.0.

Any community with shared context and strong opinions now has a way to easily make markets on anything: a fantasy league, a group chat, a Discord, a live event, a content creator with an audience.

Every opinion deserves a market.