The idea

Predict is an open-source implementation of Paradigm's pm-AMM on Solana: an automated market maker designed specifically for prediction markets.

Most prediction-market volume today runs on order books, because generic AMMs are a poor fit for outcome tokens. In November 2024, Moallemi and Robinson at Paradigm derived the AMM that fixes this: liquidity providers lose value at a uniform rate, whatever the price, and whatever the time left before the market closes.

• Paper: paradigm.xyz/writing/pm-amm
• Site: predict-pm-amm.dev
• Code: github.com/sparkfun-labs/pm-amm

A Spark idea, phase two

Spark launches ideas. Predict was one of them, and its code was built during a previous Spark hackathon.

Now Predict moves to its second phase, and it works differently from other Spark ideas: the base already exists. We're looking for people to build apps on top of it.

Already built, live on mainnet, MIT license:

• One Solana program, 26 instructions, USDC as collateral
• Binary YES/NO markets, starting at 50% or at any price from 1% to 99%, and commitment vaults that crowdfund a market's liquidity
• A TypeScript SDK on npm (@pm-amm/sdk), a full API reference, and a CLAUDE.md + llms.txt so a coding agent can build on it in one session
• 266 tests, and an on-chain solvency check on every swap

The commitment vault: liquidity from minute one

A new prediction market usually dies empty: nobody trades because there is no liquidity, and nobody provides liquidity because nobody trades. The commitment vault removes that problem.

1. Anyone opens a vault on a question, with a commit window and a minimum total.
2. People commit USDC on YES or NO. The YES/NO split sets the market's starting odds: the crowd agrees on the price before trading starts. Committing is not a bet: it funds the market.
3. Once the minimum is reached, anyone can launch the market. The whole pot is deposited as liquidity, fully collateralized.
4. Every committer becomes a liquidity provider
5. If the minimum is never reached, everyone gets refunded 1:1.

The result: a market that opens with real depth and fair starting odds, set by the people who care about the question.

What's missing is everything on top of it. Apps, front ends, communities, markets people actually want to trade. That's what this raise funds.

What you could build

The idea the Spark team most wants to see: prediction markets inside X.

Someone posts a take. Someone replies "bet?" and tags a bot. The bot opens a market on the question, starting at the odds the two sides agree on (the contract lets you set any starting price from 1% to 99%), and seeds it with liquidity. Everyone in the thread then buys their side, $10 each, and anyone else can trade it until it resolves. Every argument on X becomes a market.

Other directions, as starting points, not requirements:

• Markets between friends. The same mechanic in a Telegram group.
• Live sports. The pm-AMM works best on markets that behave like a running score (basketball, esports), exactly the case the paper was designed for.
• Infrastructure. A resolution oracle or a dispute window.

Where the fees go

Every swap pays a 2% fee, split 50% to the market creator and 50% to the protocol. The infrastructure will be updated so that the protocol's 50% flows to the Predict DAO treasury.

Why now

Solana is running Perps and Prediction Markets: $100K in prizes and one week to build perpetuals and prediction markets on Solana. Submissions close October 2.

• hackathons.solana.com/hackathons/perps-and-prediction-markets

Build on Predict for that hackathon, and pitch the same build here for the Predict treasury. The two don't conflict.

And it doesn't stop at this treasury: teams that build interesting apps on Predict can be funded as independent projects on Futard.io, with their own raise and their own treasury.

Some tips from the Spark team

• Don't rebuild the AMM. It's done. Start from the SDK and ship a product.
• Open the repo with Claude or Codex and point it at CLAUDE.md. You can have a working market on devnet in one session (mock USDC faucet included).
• Start small. One market, one use case, end to end.
• Solve resolution honestly. Today the resolution authority picks the winner. Tell users who resolves your markets and how.
• Keep it open source. Contributors with a merged pull request are listed as authors on predict-pm-amm.dev.

There is no team yet

That's the point. The engine has contributors. It doesn't have a team taking it to users.

100% of the funds are locked in a multisig. Not a single dollar unlocks without a decision market approving it. That's also why the monthly spending allowance is 0 and there is no team allocation.

Are you the builder who can take Predict to users?

Take the lead, or ship one piece of it: an app, a front end, a resolution layer, the distribution.

Come with your own terms. Tell the market what you'd build and what you want for it ($USDC? $PREDICT? Both? Investment in your own raiseon futard.io ?), then see what the market thinks. No jury, no application form, no warm intro needed.

Keep in mind:

• Distribution is the hard part. The math is solved. A prediction market lives or dies on whether people show up to trade it.
• Trust is part of the product. UX, Resolution, upgrade authority and audits are open problems (listed on the site). The team that addresses them earns the users.

How to apply / Timeline

The Spark team can open a decision market. If you want to build, ship or get involved, reach out to @mathis_btc on Telegram, or ping him in the Spark Telegram, PREDICT channel: t.me/sparkdotfun/4496.

Minimum lifetime: 1 week. If nobody has stepped up after the first week, the Spark team can submit a proposal to liquidate the treasury at any point from then on. Every token holder can then claim the treasury back. Timing is open to discussion in the Telegram, especially if people are actively working on it.

Spark's first two proposals

Spark takes no allowance and no team allocation. Our business model runs through liquidity.

Proposal 1: initialization (within minutes of launch). As soon as $PREDICT is live, Spark submits a first proposal to:

1. Redeploy the LP. Withdraw the LP position deployed on Meteora (900,000 $PREDICT) and redeploy it in a Meteora AMM with a 2.5% trading fee. Split the fees 50/50.** Half goes to the Predict DAO treasury, half to Spark.
2. Shorten decision markets from 72h to 24h, so the DAO can move at builder speed.
3. Mint 2,000,000 $PREDICT (about 15% of the current 12.9M supply) and sell them through limit orders laddered from $0.001 (the ICO price) to $0.10 per token. The tokens only sell as the price rises, adding sell-side liquidity on the way up. 80% of the proceeds go to the Predict DAO treasury, so the DAO funds itself from its own price rising, with no OTC deal. 20% go to Spark for managing the operation.

Proposal 2: sponsor the Solana hackathon. Once the first one passes, Spark submits a second proposal to put up a $1,000 Predict bounty on Perps and Prediction Markets, for the best project built on the pm-AMM.

After that, the floor is open. From the third proposal on, any builder or holder can submit theirs.

It all goes through the decision market like everything else. If the market says no, it doesn't happen.

What this is

An experiment run by Spark, and a new iteration of it. x.com/Predict_PM_AMM · x.com/JustSparkIdeas · justspark.fun

Until now, Spark funded ideas before any code existed. Predict tests the next step: fund an open-source engine that already works, and let the market pick who builds on it.

1. Fund the engine. Backers buy $PREDICT at the same price. No insiders, no pre-sale.
2. Builders compete. Teams build on the pm-AMM and pitch for the funding.
3. The market decides. A decision market (futarchy), not a jury, picks the winning team. Backers own the winner, or get a full refund if there is no winner.

More detail: justspark.fun/explanation

Disclaimer

Predict is not affiliated with, endorsed by, or supported by Paradigm, MetaDAO, futard.io or the Solana Foundation. None of them is involved in this project, its funding or its execution. The pm-AMM paper by Moallemi and Robinson (Paradigm) is credited as the mechanism this code implements, nothing more. futard.io is the launch infrastructure used here, and Perps and Prediction Markets is a public hackathon open to anyone. Nothing on this page is an endorsement from any of them, and nothing here is financial advice.

The contract on mainnet is a trust-based proof of concept: centralized resolution, a single-key upgrade authority, and no third-party audit. See predict-pm-amm.dev (§7 Limitations).