Ai Forecasting
Forecasting Models
futurefun includes on-demand AI forecasting for prediction markets. You can connect your Polymarket wallet, open any supported market, and request a forecast from one of our integrated models. Forecasts are paid per request and returned as a calibrated probability signal (and, where relevant, supporting rationale/inputs depending on the model).
How forecasting works in the app
Connect your Polymarket wallet
Open a market page on futurefun.
Click Forecast.
Select a model (availability depends on the market type).
Pay per request and receive the model’s probability estimate.
We’re a front-end and orchestration layer for multiple inference systems. Some models are experimental, and model availability depends on whether the market falls within the model’s coverage.
How to read the model output
The label shows the side the model would bet on, and the percentage is its confidence in that side. Higher confidence means lower estimated risk. This is not a guarantee, just the model’s current view.
Example
YES 78% means the model would bet on YES with 78% confidence.
NO 24% means the model would bet on NO with 24% confidence, which implies high risk.
Models:
1. AION Smart Crowd (AION)
AION Smart Crowd is our “smart crowd” signal: instead of using the average market participant, it derives a forecast from the best traders we track.
How it works (high level):
We identify a set of top-performing traders (based on performance signals we track).
For a given market, we collect their positions/exposures.
We compute an aggregate (averaging/weighting) and apply mathematical transforms to convert those positions into a probability forecast.
Coverage:
General / broad coverage: designed to work on any Polymarket market where tracked traders are active.
How to interpret it:
Think of it as “what the best of the best are collectively implying,” mapped into a clean probability number.
Status:
Testing / experimental stage. We are actively evaluating and improving the transformations and weighting logic. Use as a directional signal, not as a standalone decision engine.
2. Outcome Learned Inference (OLI)
An external outcome-reinforcement inference engine trained with reinforcement learning using verifiable rewards from real-world event resolutions. It is optimized for probabilistic forecasting (calibration + accuracy), not for creative text generation.
Training & data (technical):
Trained on a novel dataset of recent prediction questions plus relevant contemporaneous news headlines.
Reward signals come exclusively from resolved outcomes (i.e., only when events settle do they provide learning signal).
Research results in simulation suggest >10% ROI across a test set in a practical Polymarket-style trading setup, and improved probabilistic calibration with a compact ~14B reasoning model.
Inference behavior (technical):
Uses stabilized sampling / medium prediction sampling techniques at inference time to reduce noise and improve consistency across markets.
How we use it on futurefun:
This engine is used as a baseline signal inside our rotor aggregation framework.
Not proprietary to futurefun; we integrate it as a high-quality external forecasting component.
Coverage:
General markets (broad topic coverage).
3. Allora Network
Allora is an AI model coordination network: many independent ML models compete and are continuously scored and reweighted, and the network aggregates their outputs into high-confidence predictive price feeds that improve over time.
How it works (high level):
Multiple models produce forecasts for a given topic.
Models are evaluated continuously.
The network produces an aggregated forecast via dynamic weighting (higher-performing models contribute more).
How we use it on futurefun:
We integrate Allora forecasts specifically for crypto price prediction markets on Polymarket.
Coverage (important):
BTC, ETH, SOL only
Works on Polymarket markets tied to Bitcoin, Ethereum, and Solana price outcomes.
4. Synth Network
Synth delivers synthetic asset price path data generated through a decentralized network of AI models on Bittensor. This innovative approach simulates diverse market conditions, providing unparalleled forecasting accuracy and actionable insights for optimal financial decision-making.
How it works (high level):
Hundreds of AI models simulate price paths, with the most accurate forecasts rewarded through a fair and transparent mechanism.
Instead of relying solely on historical data, Synth generates real-time probabilistic price distributions, enabling dynamic risk assessment and execution strategies.
How we use it on futurefun:
We integrate Synth forecasts specifically for some crypto, commodities and equity markets on Polymarket.
Coverage (important):
Works on Polymarket markets tied to price outcomes of: BTC, ETH, XAU (Gold), SOL, SPY (S&P500), NVDA, GOOGL, TSLA, AAPL
Overall notes & limitations
Model availability depends on the market. (e.g., Allora is only for BTC/ETH/SOL markets.)
Signals are not guarantees. Forecasts are probabilistic estimates; unexpected events and thin markets can break assumptions quickly.
AION Smart Growth is experimental. Expect iteration and improvements over time.
Want more models?
We’re expanding coverage and adding new inference integrations as we validate performance and calibration. If you have a model you want us to support, or a market category you care about, reach out to [email protected] and we’ll prioritize based on demand.
Last updated
Was this helpful?