This is a high level overview covering some key parts of the Autonity ecosystem. Future posts may include walk-throughs or guides.

As mentioned in previous articles, Autonity is a new public EVM-compatible blockchain built specifically for decentralized clearing of futures contracts. In plain terms, it’s a blockchain made to run and settle smart futures contracts using real-world data such as inflation, unemployment, and GDP.

Autonity has two native tokens: ATN for gas (transaction fees) and NTN for staking (securing the network). Its mainnet launched on August 12, 2025. Before mainnet, the community ran several incentivized testnets (Piccadilly and Tiber) where participants tested the system with various competitions. The Bakerloo testnet is live and mirrors mainnet’s production settings.

Getting Started (Wallet, Tokens, and Bridges)

To use Autonity, first set up a Web3 wallet (like MetaMask or Coinbase Wallet) in your browser. Then add the Autonity network to it. You can use Chainlist.org or manually enter the network details (Chain ID 65000000).

You’ll need ATN in your wallet to pay for gas. ATN for mainnet can be acquired by bridging (via Proto USD bridge by VIA Labs) USDC from other chains (like Polygon) into Autonity or by swapping in an AMM.

During the testnet competitions there were several AMMs. As mentioned in the Discord, there are a couple of Uniswap clones deployed. Most without frontends yet.

On mainnet, a UniV2 clone:

• Factory: 0x4D570019De75c2488740B9F153953802dDdda0Be
• Router: 0xdAC5261966B2Cd26C220f4F4fDB1D7A9468F4BF0

On mainnet, a UniV3 clone:

• Factory: 0x640C1478e1261CC07a804006aBEEa0F657Fd001e
• NonfungiblePositionManager: 0xB55c5C3B1043352C10e5CDD1F0f92dA24e5a8711
• SwapRouter02: 0xEEE015aFC3a12d330cB9Ee603D1F11FDd320848a
• QuoterV2: 0x8f85b46133a4D55372BCf212aB012A02803F9b45

On Bakerloo testnet, a UniV2 clone:

• Factory deployed at: 0x9709D1709bDE7C59716FE74D3EEad0b1f12D3944
• Router deployed at: 0x13a3a74463218D123596386D3E36bd1aC13DCFE2

On Bakerloo testnet, a UniV3 clone:

• UniswapV3Factory: 0xCa5df1F426Db6cAE41B8E22404934BAfa986532f
• SwapRouter: 0x6A613f3aC76eEA735F92911522FE4D6B4B1dAd4f
• WATN: 0x7152e69E173D631ee7B8df89b98fd25decb7263D
• USDCx deployed at: 0x90488152F52e1aDc63CaA2CDb6Ad84F3AEC1df3E

Anyone can obtain ATN for the Bakerloo testnet via a (daily) faucet via FaucetMe.

Note also that during the Forecastathon competition, Autonity uses a fake stablecoin USDz for collateral) at address (0xAB436b), which you can add to your wallet to see your balance.

Autonomous Futures Protocol (AFP)

At Autonity’s core is the Autonomous Futures Protocol (AFP) – an on-chain clearing system for futures contracts on any data series. Unlike typical DeFi futures, AFP decouples trading from clearing: any exchange or trading platform can plug in, and all trades share a common on-chain margin account system (see the DCC paper). In practice, AFP lets anyone create a new futures product (e.g. “US CPI Oct 2025”) that settles on real-world data. When you trade, your positions and collateral are managed in a public on-chain margin account within AFP. All of AFP’s logic (netting, mark-to-market pricing, auto-liquidation) runs on Autonity, so it’s transparent and automatic and not tied to a specific trading venue.

Getting and Funding an AFP Account: To trade on AFP, you need collateral. For the Forecastathon, collateral is automatically provided as: 100,000 “play” USDz, plus a little bit of ATN for gas. Once funded, the user deposits USDz into their AFP margin account:

1. Visit the Forecastathon site: Go to forecastathon.ai/margin-account and click “Connect Wallet.” Use the same wallet you registered with.
2. Sign in (gas-less): A pop-up will ask you to sign a (gasless) message. This just proves wallet ownership.
3. Deposit collateral: Click Deposit on the USDz row, enter an amount (e.g. all 100,000 USDz), and confirm. (You’ll pay a small ATN fee for this on-chain transaction.)

After depositing, your on-chain AFP margin account is ready and appears on the site. You can deposit more USDz later if needed; the account is managed by a smart contract on Autonity.

Once the Forecastathon is completed users can continue to use ATN or USDC as collateral.

Trading on Autex (On-Chain CLOB)

The first trading venue for AFP is Autex, an on-chain central limit order book (CLOB). Autex is a browser-based exchange (at autex.exchange) that connects to AFP. Here’s how to trade:

1. Choose a market: In your connected wallet, go to autex.exchange and open the Products list. You’ll see futures like US Unemployment rate, US GDP, US CPI. (These are “Forecast Futures” that settle on data.)
2. Select a contract: Pick a product and month, then provide a “BUY” or “SELL” limit order. Enter the quantity of contracts and the price you want (in USDz). The UI will show how much initial margin (IM) is needed; make sure your available margin (USDz in the account) is ≥ IM.
3. Sign your order (gaslessly): When you submit the order, your wallet will pop up to sign it. This signature is gas-free: it’s only a cryptographic proof of intent, not a blockchain transaction. The actual trade data is sent to AFP on-chain by the matching engine.
4. Order management: Your open orders show up under “Open Orders” until they fill. When an order executes, the on-chain clearing system transfers collateral and PnL automatically. You can view your positions and profit/loss on Autex (under “Positions” and “Order History”). Any order execution details (with on-chain tx IDs) appear in history.

All trading on Autex occurs on Autonity mainnet (the same final ledger and data as any other chain transaction). During the first half of the Forecastathon, the Autex was the sole exchange, but the AFP design allows other venues to launch later. Season 1 of the Forecastathon saw 275,000+ trades on Autex.

Exploring and Monitoring Data

• AutonityScan: a block explorer where you can browse blocks, transactions, addresses, and contracts. This is a BlockScout-based explorer just like Etherscan, adapted for Autonity’s Mainnet. You can search any TX hash or wallet to see on-chain history.
• AFP Explorer: The AFP Explorer is a special front end for AFP data. It works like a block explorer for futures products: you can lookup any AFP product (by its contract) or your margin account to see positions and history (this is independent of the Autex). It displays settlement prices, open interest, trades, and futures products.
• Autex Products List: You can also see tradable products directly on Autex’s interface. The /products page lists all active contracts (e.g. CPI, Unemployment, GDP) with ticker codes and next data date.
• Stakeflow (Validators & Oracles): For network health and validator info, use the Stakeflow Autonity page. Stakeflow is a multi-chain explorer that shows the active validators, their total stake, commission, uptime, and even oracle data feeds. (For example, it shows which price oracles each validator runs.) Viewing a validator’s profile will show identity info and commission rates. Stakeflow makes it easy to research validators before staking.
• Liquidity Pools: Leibniz and other AMMs (mentioned above)

Command-Line Tools (CLI)

Lastly, for developers or power users, Autonity offers a command-line interface Autonity CLI (the aut tool). This Python-based CLI (on GitHub) lets you do everything from the terminal. With it you can create accounts, check balances of ATN or ERC-20 tokens, send transactions, deploy or call smart contracts, query block data, and even manage staking and validator operations. It connects directly to any Autonity JSON-RPC node. To install, you use pipx install autonity-cli and then run commands like aut transfer, or aut query-block. (See the README on GitHub for details.)

The next post will take another look at comparing the AFP (and forecast futures) with other existing infrastructure categories.

This post will briefly compare and contrast the AFP with a few well-known perpetual DEXes.

What is the AFP?

As mentioned in the previous articles, Autonity is an Ethereum‐compatible (EVM) blockchain built specifically for decentralized derivatives clearing. Its Autonomous Futures Protocol (AFP) runs on Autonity as a clearing layer for futures contracts. In AFP, anyone can create a “Forecast Future” product based on any real‐world time series (like GDP, CPI, or jobless claims) and link it to an oracle price feed. During the current trading competition, traders deposit a stablecoin (USDz) into a margin account and then use a front-end exchange, AutEx (a non-custodial central limit order book) to place buy/sell orders. When an order is filled, the trade is sent on-chain to the AFP clearing system, which verifies and settles the trade.

AFP’s key features are decentralized clearing, cross-venue margining, and futures on any timeseries. “Cross-venue margining” means that all collateral sits in one on-chain account, so profits from one market (say an inflation future) automatically add to your balance for another market. This design separates trading from clearing: any trading interface (orderbook or AMM) can integrate with AFP. The current trading contest (“Forecastathon”) uses the AFP to trade macro futures – for example “U.S. CPI Oct 2025” futures settled on published CPI data.

In sum, AFP is a permissionless clearinghouse: it records every trade on Autonity’s chain and can settle dated futures on arbitrary data, without a central intermediary.

Hyperliquid

Hyperliquid is a perpetual futures exchange built on its own high-performance Layer-1 blockchain. Unlike AFP, Hyperliquid is self-contained: its blockchain (called HyperEVM) runs an on-chain matching engine. Users deposit crypto (e.g. USDC) into Hyperliquid and trade directly on its orderbook. Hyperliquid’s goal is CEX-like speed: it uses a special consensus (“HyperBFT”) with around 20 validators to achieve sub-second finality and ~200,000 transactions per second. Because of this throughput, Hyperliquid can run a full limit-order book on-chain and offer nearly gas-free trading. In practice, traders enjoy “1-click” buy/sell of perpetual futures with extremely low fees.

Hyperliquid combines features of both centralized and decentralized exchanges. It is permissionless (no central custody of user keys), but it feels like a CEX in speed and interface. Every order, trade, and cancellation is recorded on its blockchain, so all activity is transparent. Hyperliquid has its own native token (HYPE) and a “liquidity vault” mechanism (HLP) to support deep liquidity. In short, Hyperliquid is a high-speed L1 DEX with an on-chain CLOB, focused on crypto perpetuals.

dYdX (v4 on Cosmos)

dYdX is a long-standing decentralized exchange that has moved to a new architecture (v4). The dYdX team has built its own Cosmos-based blockchain (dYdX Chain) that runs a fully decentralized off-chain orderbook. In this system, every validator node keeps an in-memory copy of the orderbook. Traders submit orders (off-chain) and validators match them together; only the resulting trades (fills) are written to the chain each block.

Key points about dYdX v4:

• It is a standalone Tendermint/PoS chain (built with Cosmos SDK). This means it has its own token, validators, and is independent of Ethereum (it’s not an L2).
• The orderbook and matching are decentralized and off-chain. Because validators handle matching, dYdX can scale to very high throughput without central matchers.
• Traders do not pay gas fees on each order. Instead, they pay trading fees (like a CEX) that go to validators.
• The chain is open-source and permissionless; no component is controlled by dYdX affiliates.

In practice, dYdX supports perpetual futures on crypto assets. Its design shares similarities with AFP (on-chain settlement) and Hyperliquid (high-performance orderbook) but mixes them differently. All trades (deposits, withdrawals, fills) happen via transactions on the dYdX Chain, but actual matching happens off-chain by validators for speed.

Lighter

Lighter is a decentralized perpetuals exchange built as a zero-knowledge (zk) rollup on Ethereum. It operates on a custom ZK infrastructure (currently on Arbitrum) to run a CLOB-style DEX with cryptographic guarantees. Users keep their funds in Ethereum contracts, and a sequencer processes batches of orders with very high throughput.

Important features of Lighter:

• ZK-proofs for trades: Every order match, liquidation, and funding calculation is turned into a zk-SNARK proof that gets posted to Ethereum. This means the correctness of the exchange’s operations is publicly verifiable on Ethereum.
• High performance: Lighter claims to process “tens of thousands of orders per second” with millisecond latency. It achieves this by batching and proving off-chain.
• Fee structure: Lighter offers zero trading fees for retail users via the web UI; it only charges fees for API and high-frequency usage. (Gas fees on Ethereum are only paid when users deposit/withdraw.)
• Security: Because Lighter is a zk-rollup, security is ultimately based on Ethereum. If the sequencer stops, users can exit via proof or an emergency mode.

In summary, Lighter is an Ethereum layer-2 (zk-rollup) DEX for crypto perpetuals. It has a central sequencer (for speed) but uses ZK cryptography to guarantee trustlessness and fairness.

Similarities

• Decentralized futures trading: All four platforms let users trade futures/perpetual contracts without a central exchange holding custody. Trades are ultimately settled on blockchain, ensuring transparency.
• Leverage and margin: Each platform uses collateralized margin accounts so traders can use leverage. For example, AFP traders currently deposit USDz to open futures, and Hyperliquid/dYdX/Lighter traders deposit USDC or other tokens to margin positions.
• No per-trade gas for users: In practice, trades feel “gasless” to traders on all four. Hyperliquid’s chain subsidizes gas (zero fees for trades), dYdX’s Cosmos chain does the same, Lighter’s fees are waived for retail, and AFP’s trades via AutEx during the Forecastathon are gasless for users (the foundation temporarily covers clearing gas).
• High-speed focus: Each is designed for high throughput. Hyperliquid boasts 200k TPS, Lighter does thousands/sec with ZK batching, dYdX v4 can scale via distributed validators, and Autonity’s Tendermint consensus gives sub-second finality for AFP.
• Advanced orderbooks: They all use orderbook trading rather than simple AMMs. Hyperliquid and Lighter run full CLOBs, dYdX uses a high-end off-chain book, and AFP’s AutEx is also a non-custodial CLOB. This suits professional trading (tighter spreads, complex orders).
• Permissionless architecture: Each platform is permissionless/open-source. Anyone can set up a node or interface. AFP even lets any user create new futures products. All use oracle data for prices.
• Crypto settlement: except for AFP’s macro futures, all settle on crypto prices. Hyperliquid, dYdX, and Lighter focus on crypto (and sometimes tokenized equities), while AFP’s forecast futures settle on real-world data like inflation.

Differences

• Underlying technology: AFP runs on the Autonity L1 (Ethereum-based PoS chain) that is separate and independent from trading venues. Hyperliquid runs on its own HyperEVM L1 with built-in order matching. dYdX v4 is a Cosmos L1 where validators match orders off-chain. Lighter is a Layer-2 zk-rollup on Ethereum. In short: AFP = new EVM chain + separate exchange, Hyperliquid = bespoke L1 chain with DEX, dYdX = Cosmos chain + validator network, Lighter = Ethereum zk-rollup.
• Product type: AFP’s futures currently have set expiry dates and can be on any data series (e.g. U.S. GDP, energy demand). The others only offer perpetual contracts (no expiry) on assets like BTC, ETH, or stocks. This makes AFP unique in scope (macro data vs. crypto tokens).
• Order matching: Hyperliquid and Lighter handle matching on-chain (Hyperliquid via its own chain, Lighter via proofs on Ethereum). dYdX handles matching off-chain by validators. AFP (currently via AutEx) collects orders through an intents-based model and then every filled trade is cleared on-chain.
• Consensus and decentralization: Hyperliquid’s chain is “relatively centralized” (only ~16 validators) for speed. dYdX’s Cosmos chain will be fully decentralized (potentially hundreds of validators). Lighter uses Ethereum’s security but has one sequencer (with exit options). Autonity uses Tendermint BFT PoS (similar to Cosmos) for finality and more than 25 independent validators.
• Collateral/token: for the competition AFP currently uses USDz (a stablecoin) and its native gas token is ATN. Hyperliquid uses HYPE and stablecoins (USDC). dYdX v4 uses its DYDX token (with USDC collateral). Lighter uses stablecoins/ETH and recently launched a LIGHT token.
• Fees and economics: All four have no trading gas for users, but they differ in tokenomics. Hyperliquid funds trade fees into HYPE buybacks, Lighter waives fees for UI users, dYdX uses staking and trade fees to reward validators, and during the trading competition the foundation pays clearing gas on behalf of traders.
• Unique features: AFP’s standout feature is cross-venue cross-margining (one margin account for all trades). Hyperliquid’s is raw speed and a built-in vault (HLP) for liquidity. dYdX’s is Cosmos-centered. Lighter’s is zk-proof guarantees and zero-fee trading.

The next posts will look more closely at the Forecastathon and other elements of Autonity’s infrastructure.

This post briefly explores the differences and similarities of the Autonity / AFP forecast markets with the existing communities of: Kaggle, Metaculus, Numerai, and Manifold Markets.

AFP

As mentioned in the previous article, Autonity’s Autonomous Futures Protocol (AFP) is a new blockchain-based system for creating and trading forecast futures, special contracts tied to real-world data.

For example, AFP lets people trade contracts whose value depends on future data like the U.S. Consumer Price Index or jobless claims. Instead of simple yes/no bets, these contracts pay out continuously based on the final data value (e.g. the exact CPI number).

The AFP is built on Autonity (an EVM Layer-1 chain) and uses a novel design that decouples trading from clearing. In plain terms, this means any trading venue connected to AFP can let users buy or sell the same futures contracts, while all the risk/margin is handled by a shared clearing system. An Autonity blog post explains that AFP “consolidates margin and open interest across venues” to pool liquidity and capital efficiency.

In other words, markets aren’t siloed on one website. AFP’s first live trading venue is called Autex, and it has launched prediction contracts on weekly U.S. initial jobless claims and monthly U.S. CPI (inflation) data. These are called forecast futures markets, a new category of derivative where predictions become tradable assets.

The AFP’s design is very different from typical prediction sites. Instead of just betting on a binary event, traders can take positions on any continuous outcome (like a precise economic number). AFP is decentralized and uses real crypto assets (on Autonity) for trading. It is also open and permissionless: anyone can create a new forecast future on any time series data. Because of this, the AFP community sees it as a bridge between models and markets, turning statistical forecasts into tradable financial instruments.

Manifold Markets

Manifold Markets is an online prediction market platform focused on social forecasting. On Manifold, users create and trade markets about upcoming events (like elections or sports), using play money called Mana. For example, a question might be “Will the next U.S. President be a Republican?” and users bet Mana on “Yes” or “No”. Manifold’s markets use an automated market maker (Maniswap) under the hood to set prices. It was originally called Manifold Markets and even briefly allowed real-money bets (a feature called Sweepcash, now discontinued).

Similarities to AFP: Both Manifold and AFP are communities built around forecasting the future. Each allows people to make predictions about real-world events and see if their insight was right. They both create a “market” for predictions, in the sense that participants can gain or lose based on what happens. In that way, they encourage data-driven thinking and let users test their forecasting skill.

Differences from AFP: There are several big differences.

First, Manifold’s bets are binary or categorical: you typically bet on one of two (or a few) outcomes. AFP’s forecast futures are continuous contracts tied to numerical data (like a CPI value), not just yes/no questions.

Second, Manifold is a single web platform (centralized), so each market lives only on Manifold’s site. AFP markets, by contrast, can be traded on any exchange that integrates AFP (currently Autex, and potentially others). This means AFP markets share liquidity across venues, whereas Manifold markets are isolated.

Third, Manifold uses play money (Mana) with no real financial risk (trading Mana is just for fun and reputation). AFP trades use actual assets (ATN and USDC) and collateral on a blockchain, so real value is at stake.

Finally, Manifold’s community is more casual and game-like (with users casually betting on sports, politics, or tech news), while AFP is targeting quant traders and data enthusiasts who want to trade real derivatives on economic or scientific data.

• Market type: Manifold is a centralized play-money prediction market; AFP is a decentralized blockchain futures market.
• Outcomes: Manifold bets on discrete events (e.g. yes/no questions); AFP contracts settle on continuous time-series values (e.g. inflation numbers).
• Trading: Manifold’s trades stay on its own site. AFP trades happen on Autonity’s network and can occur on any connected platform (no silo).
• Currency: Manifold uses in-platform Mana (no real money risk); AFP uses real crypto assets on a Layer-1 chain (real economic stakes).

Metaculus

Metaculus is a long-running forecasting platform and community. Users on Metaculus make probability or numeric predictions about a wide range of questions, especially scientific, technological, and global trends. Metaculus features three kinds of forecasts: yes/no (binary) questions, numerical-range forecasts (give a range of a quantity), and date-range forecasts (guess when something will happen). The community submits forecasts, and Metaculus aggregates them into a consensus probability. Forecasters earn “points” for accurate predictions, and there are reputation and prize systems for top performers. Metaculus is notable for focusing on data and analysis; users can even suggest new questions, which (after moderation) get opened to everyone.

Similarities to AFP: Like AFP, Metaculus deals with forecasting real-world data and events. Both involve users making predictions about measurable outcomes. Metaculus also goes beyond binary questions by allowing numeric forecasts for things like “What will next year’s world population be?” or climate questions.

Differences from AFP: The main difference is that Metaculus is not a trading market – it’s a prediction aggregation site. Users on Metaculus contribute forecasts to a community tally; there is no actual buying or selling of contracts. There is no shared liquidity or venue integration: Metaculus forecasts stay on Metaculus. AFP, by contrast, turns each forecast into a financial derivative that can be traded for profit or loss. Metaculus uses abstract points and reputation to reward accuracy, whereas AFP uses real assets to reward gains and losses from trading. Also, Metaculus questions are usually created by the site (often academic/science topics) and chosen by a moderator team, while AFP is permissionless – anyone can spin up a new forecast market on any allowed data series. Finally, Metaculus forecasts are often about when or what specific threshold something will happen (like “Will X reach Y by year Z?”), whereas AFP’s forecast futures are directly settled on the outcome number (e.g. the exact CPI value on a given date).

• Market type: Metaculus is a centralized forecasting site (no financial market); AFP is a decentralized trading market.
• Format: Metaculus aggregates user probability/numeric forecasts and scores accuracy; AFP offers actual futures contracts that pay out based on real data.
• Currency: Metaculus uses points and occasional cash prizes for top forecasters; AFP uses crypto assets in open markets.
• Community: Metaculus is research-oriented, focusing on science/tech questions; AFP is aimed at traders and quant-modelers who want to trade derivative products.

Numerai

Numerai is a decentralized hedge fund and data science network. It runs regular data modeling competitions: data scientists worldwide download anonymous stock-market data, build machine-learning models, and submit predictions to Numerai’s weekly tournaments. Numerai then uses the combined winning models to trade a real hedge fund. Participants stake crypto (Numeraire, NMR) on their models: if their prediction is good, they earn NMR; if it’s wrong, their stake is lost (burned). Numerai famously calls itself “the hardest data science tournament on the planet” because it focuses on complex stock-return predictions. It has paid over tens of millions in NMR rewards to its community.

Similarities to AFP: Both Numerai and AFP blend forecasting with cryptocurrency and market incentives. In both communities, participants make predictions (in Numerai’s case via ML models, in AFP’s case via trades and forecasts) and can earn real token rewards. Both are “prediction games” with financial stakes tied to accuracy. Also, both appeal to quants and data enthusiasts who use models to forecast numeric outcomes.

Differences from AFP: Numerai is quite different in structure. It is not a public market – there is no open buying/selling of contracts on a marketplace. Instead, it is a closed prediction tournament specifically about stock-market data. Numerai provides data and takes in models; AFP provides markets that anyone can trade. Numerai’s participants never directly trade assets; they just submit forecasts and stake on them. AFP participants trade contracts and keep or lose money directly from prices. Numerai uses an internal token (NMR) for staking and focuses narrowly on equity returns, while AFP uses the Autonity blockchain’s assets and can be used for any agreed-upon time series (inflation, climate data, crypto metrics, etc.). In short, Numerai is a crowdsourced hedge fund contest, whereas AFP is a marketplace for forecast contracts.

• Platform: Numerai is a private modeling competition tied to a hedge fund; AFP is an open futures market platform.
• Predictions: Numerai users submit AI model forecasts on stock data (weekly); AFP users trade contracts on economic indicators (continuous).
• Incentives: Numerai uses staking of its own crypto token (NMR) to reward accurate models; AFP uses trading profits/losses on real tokens.
• Community: Numerai’s crowd is mostly data scientists working on finance problems; AFP’s crowd can include traders, students, or anyone interested in quantitative forecasting.

Kaggle

Kaggle is a well-known platform for data science and machine learning competitions. Organizations and researchers post datasets and challenge participants to build the best models (for example, image recognition, predicting house prices, or time-series forecasts). Competitors submit code or predictions, and Kaggle scores them, often with cash prizes or recognition for the top teams. Kaggle also hosts community tools (like shared code notebooks), but fundamentally it is a contest arena for solving data problems.

Similarities to AFP: Kaggle and AFP both involve forecasting and modeling data. A Kaggle competition could be about predicting next month’s economic numbers, which is similar in theme to an AFP market on the same data. Both communities encourage improving prediction accuracy and sharing techniques. Also, both platforms sometimes attract students, academics, and tech-savvy users who enjoy applied data challenges.

Differences from AFP: The differences are substantial. Kaggle competitions are offline contests, not real-time markets. Participants submit their model outputs after seeing fixed datasets, and scoring is done once per competition. There is no continuous trading or live market dynamics. In contrast, AFP is a live market where anyone can buy/sell contracts at any time before settlement. Kaggle does not involve financial trading or tokens (aside from contest prizes); it’s mostly about reputation and winning contest money. Moreover, Kaggle’s focus is broad (image analysis, NLP, forecasting, etc.) and usually one-time challenges, whereas AFP focuses specifically on prediction markets tied to time-series outcomes. In short, Kaggle is about building the best static model for a given dataset; AFP is about trading dynamic contracts based on ongoing data releases.

• Activity: Kaggle runs batch-style competitions on posted data; AFP runs live trading markets on time-series.
• Output: Kaggle outcomes are model submissions and final rankings; AFP outcomes are contract prices and P/L.
• Currency: Kaggle uses points and prize money for contests; AFP uses blockchain tokens in open markets.
• Flexibility: Kaggle has organizers who set each problem; AFP allows any user to propose a new market on any permissible data.

Summary

All five platforms – AFP, Manifold, Metaculus, Numerai, and Kaggle – build communities around predicting the future, but they do so in very different ways:

• Market vs Contest: Manifold and AFP are markets where predictions are traded, while Metaculus and Kaggle are more like contests or forums for submitting forecasts. Numerai is a hybrid: a contest that feeds into a real hedge fund.
• Real Money vs Play: AFP (and Numerai) involve real crypto at stake, whereas Manifold and Metaculus mostly use play-money or points (Manifold’s Mana, Metaculus’s points). Kaggle uses prize money only at contest end.
• Continuity: AFP markets are continuous and integrated across venues. The others generally have separate, siloed events or questions.
• Data vs Models: Kaggle and Numerai emphasize building models on datasets. AFP and Manifold emphasize trading predictions (though both can require modeling skill). Metaculus emphasizes individual probability estimates without trading.

In short, all these communities share the fun of forecasting, but AFP’s twist is turning each prediction into an actual financial futures contract on a blockchain, tradable across platforms. This creates a new kind of engagement: instead of just guessing or coding offline, participants can trade their forecasts in real-time markets. The other platforms either lack this trading element or focus on different domains (binary events, science forecasts, machine learning contests).

In summary, AFP is similar to these communities in its goal of leveraging crowds to forecast data, but differs by using decentralized finance mechanics and time-series derivatives to make those forecasts tradeable across venues.

The next post will dive into the world of “decentralized perpetuals.”

This article is a simple, non-comprehensive comparison between two existing prediction markets up against a new category maker (forecast markets).

The Autonomous Futures Protocol (AFP) is a new decentralized platform built on the Autonity blockchain (an EVM-compatible Layer-1). The AFP supports “Forecast Futures” – futures-style contracts whose payoff depends on real-world time‑series data (like CPI or unemployment) rather than a simple binary yes/no event.

For example, current Forecastathon markets include weekly U.S. jobless claims and monthly U.S. CPI forecasts, settling to the actual reported data. AFP’s core architecture decouples trading from clearing: all trades are ultimately settled on a shared on‑chain clearing system, even though anyone can build trading venues that plug into AFP. In practice, AFP defines margin accounts and clearing rules on-chain, and any compatible exchange (called a trading protocol) can match orders.

The first such venue, AutEx, is a non‑custodial order-book exchange connecting to AFP’s clearing contracts (AutEx is expected to be open-sourced soon so others can deploy similar exchanges). This design means a product is “created once, traded anywhere.” This means a Forecast Future created on AFP can be traded on AutEx or any other AFP‑integrated venue, sharing liquidity and collateral. Traders deposit collateral (e.g. USDC or ATN) into on-chain margin accounts and can use unrealized gains from one venue as margin on another (cross-venue cross‑margining). Product creation on AFP is permissionless – any individual can register a new time-series contract as long as it meets data-source requirements.

In summary, AFP is a blockchain‑cleared derivatives framework for continuous outcome markets, with multiple trading venues and shared clearing (no single “house” or silo).

How it is built

AFP runs on Autonity (an EVM-based L1). Its smart contracts manage margin accounts, oracles, liquidation, and settlement. It separates trade execution (the matching of orders on an exchange) from clearing (updating on-chain accounts), removing the monopoly vertical silo by enabling new venues to plug-in.

Product design

AFP’s “Forecast Futures” are continuous futures on time-series (not binary yes/no). Each contract is defined by a dataset (e.g. “US monthly CPI”) and a date. Settlements come from trusted oracles. The payoff can be any function of the data (for example, it could pay more for higher inflation values). This lets markets capture the full range of possible outcomes. The first markets (Forecastathon) rewarded accurate forecasts of real economic numbers, aiming to aggregate predictive models.

Polymarket

Even though they are frequently in the news, if you aren’t familiar with them: Polymarket is a decentralized prediction market platform built on the Polygon blockchain. It allows users to wager on the outcome of future events (yes/no questions) using crypto. In Polymarket, each market asks a binary question (e.g. “Will Candidate X win the election?”). There are two tokens (YES and NO) worth $1 if that outcome happens, $0 otherwise. Polymarket uses USDC (a stablecoin) as collateral: every pair of YES/NO shares is fully collateralized by $1 USDC. In effect, if you buy a YES share for 60¢, someone else is effectively selling a NO share for 40¢, and when the market resolves the $1 is paid out to the winning side.

Kalshi

Kalshi is a centralized, regulated exchange (a CFTC‑licensed Designated Contract Market) for trading yes/no event contracts. Like Polymarket, Kalshi’s products are binary questions about future events, but its structure is very different: it operates as a traditional exchange platform under U.S. regulation.

With that brief introduction, let’s look at some key comparisons.

Underlying Tech

AFP: Built on the Autonity L1 blockchain. Clearing logic and margin accounts on-chain. Open “shared ledger” where multiple trading venues (like AutEx) can plug in

Polymarket: Deployed on Polygon. Uses on-chain smart contracts for each market. Similar to the AFP, it is fully decentralized – users’ funds remain in smart contracts (non-custodial).

Kalshi: Centralized proprietary platform (backend servers and database). Operates under CFTC oversight as a futures exchange. User funds are held in custody by Kalshi.

Type of Contracts

AFP: Forecast Futures on continuous data. Contract outcome is a function of a numeric time-series (e.g. next month’s CPI index). Payoff can vary over a range of values. These are derivatives, similar to futures on data streams.

Polymarket: Binary prediction (yes/no) markets. Shares are priced $0–$1, effectively betting on a probability. Only two outcomes per market.

Kalshi: Binary event contracts. Exactly like Polymarket, each market has YES/NO, paying $1 if correct. No continuous or multi-outcome features.

Market Creation

AFP: Permissionless creation. Any user (a product builder) can define a new time-series product, register it on-chain, and start trading it. There is no central approval needed (no “governance gating”). Once registered and listed on a venue, it becomes live.

Polymarket: Curated by team. Users cannot directly deploy their own markets on-chain. Instead, Polymarket’s staff (with community suggestions) decide which questions to list. This is a semi-centralized process – markets must fit Polymarket’s guidelines.

Kalshi: Company-controlled. Kalshi itself creates all markets (often with input from customers or in response to regulation). As a DCM, Kalshi must ensure each contract meets legal standards (e.g. approved event definitions). There is no user-driven market creation.

Trading & Venues

AFP: Multi-venue. Once a product exists, it can be traded on any AFP-compatible exchange. Trades from all venues clear in the same system, enabling one common price and shared liquidity across venues. For example, a bid on AutEx could match with an order on a future exchange that also uses AFP. Collateral can move freely, and profits anywhere act as margin anywhere.

Polymarket: Single venue. Markets can only be traded on Polymarket itself (and technically on any chain where its contracts are deployed, but in practice Polymarket is its own DApp). There is no way to trade a Polymarket market on another platform. Liquidity and order books are confined to Polymarket’s ecosystem.

Kalshi: Single venue. All trading happens on Kalshi’s exchange. Its contracts cannot be accessed or traded elsewhere. Liquidity is limited to Kalshi’s user base. No cross-platform trading is possible.

Collateral & Clearing

AFP: Uses blockchain-based margin accounts. Traders deposit assets such as USDC or ATN into on-chain accounts. Positions can be leveraged using margin, with rules for liquidation if undercollateralized. Because clearing is on-chain, AFP supports cross-margining across venues.

Polymarket: Uses USDC (on-chain). No leverage – you can only bet the USDC you deposited. Trade settlement and payout are handled by smart contracts atomically (no separate clearinghouse). If your shares win, you get $1 per share back from the contract. There are effectively no borrowing or margin positions.

Kalshi: Uses deposited USD (or stablecoins). Like Polymarket, positions are cash‑collateralized 1:1 (no credit). Kalshi manages an internal clearing ledger under regulatory rules. Traders simply have fiat balances that gain or lose $1 per contract; Kalshi handles payouts from its pool of collateral.

Example Products

AFP: Publicly launched ones include “U.S. CPI for October” or “Weekly Jobless Claims”. These pay off according to actual reported numbers.

Polymarket: Questions like “Will candidate X win?” or “Will Bitcoin exceed $50k by year-end?”. These resolve to true/false.

Kalshi: Events like “Will GDP grow above 2% this quarter?” or “Will Fed raise rates?” Again yes/no questions, but only listed with Kalshi’s approval.

Decentralization vs. Regulation

AFP: Fully decentralized clearing on a public blockchain. Anyone worldwide can (in principle) connect and trade without permission.

Polymarket: Decentralized DApp (no central authority). In practice, Polymarket excludes U.S. users due to law, and uses an on-chain Oracle (UMA) for disputes. But users keep control of funds.

Kalshi: Centralized and regulated. Trades require an account, KYC, and follow U.S. futures regulations. The CFTC oversees Kalshi’s rules. Non-U.S. residents may also trade, but it is a traditional “walled garden” exchange.

Summary

In summary there are just a few similarities, such as how both Polymarket and the AFP use smart contracts on a blockchain, but their contract types are very different and how the contracts are traded are even further apart. Architecturally they are both designed to solve different problems: prediction markets such as Polymarket attempt to solve binary outcomes on discrete events whereas the AFP allows forecasters to trade on arbitrary, reoccurring time series.

The next couple of articles will look at how the AFP fits in the world of other communities and ecosystems.

We started the year primarily focused on the type of long-form, human readable content that a couple SOTA LLMs were able to create with their agentic tools.

Throughout the summer we became increasingly interested in the world of forecast markets, such as the communities of Kaggle, Numerai, Manifold, and Metaculus.

We think there is a bit of alpha that can be gained by sifting through not just the data sets and methods of the data scientists and quants that participate in their polls, contests, and contracts, but that there is a clear intersection with data notation, training, and deploying forecast models for new platforms and venues.

Over the coming months we will dive into a new protocol we think captures that zeitgeist: Autonity.