Core Vision: Enable global predictive models to continuously evolve through open-market competition, validated by real-world outcomes, producing ever-improving event simulation and forecasting capabilities.
1. Introduction: Decentralized Event Simulation & Prediction Engine
MiroSwarm is a Bittensor subnet dedicated to building a decentralized, continuously evolving event simulation and prediction system.
The Problem: The AI prediction market today suffers from three fundamental flaws:
- Single-methodology lock-in: Platforms rely on a narrow set of models or analysts.
- No real-world validation loop: Predictions are rarely tracked and scored against actual outcomes.
- No scalable incentive for quality: The best forecasting capabilities cannot be reliably identified, rewarded, or scaled.
MiroSwarm's Solution: We commoditize "predictive intelligence" on Bittensor. Validators feed live real-world events to miners, who generate structured forecasts across three time horizons (2-day, 7-day, 14-day). After the fact, validators collect ground-truth outcomes and objectively score each miner against their predictions. Yuma Consensus allocates TAO emissions based on these scores, creating a self-evolving prediction engine.
Key Figures:
- 3 validation windows (2-day / 7-day / 14-day) covering short, medium, and long-range forecasting.
- N+ competing methodologies (geopolitical models, market-sentiment models, tech-diffusion models, social-network models).
- $XXB+ global predictive analytics market (finance, geopolitical risk, policy impact assessment).
2. Reward Logic
MiroSwarm employs a Benchmark-Beating + Tiered hybrid reward structure.
Core Principles:
- Breakthrough Reward: A miner that sets a new accuracy benchmark captures the majority of miner-side emissions for that cycle (up to 80%).
- Sustained Competitiveness: Strong contenders near the benchmark share the remaining emissions proportionally.
- Elimination: Miners consistently ranked at the bottom lose their UID slots to new registrants.
Emission Flow (standard Bittensor split):