Stephen Rothwell: Sportstensor, Polymarket, Bittensor Subnet 41, Signal Markets, Incentives | Ep. 69 скачать в хорошем качестве

Stephen Rothwell: Sportstensor, Polymarket, Bittensor Subnet 41, Signal Markets, Incentives | Ep. 69

In this episode we are joined by Stephen Rothwell, Head of AI at Sportstensor, operators of Bittensor Subnet 41. Stephen returns for his second appearance to dive deep into their groundbreaking partnership with Polymarket, the hostile landscape for signal providers in traditional betting, and how Subnet 41 creates a meta-market for high-signal inference. He breaks down price vs. signal efficiency, incentive mechanisms, data challenges, and the path to training ground truth in prediction markets. In this episode, you’ll learn Stephen’s perspective on the $2.4T global betting market and why traditional bookmakers ban or limit sharp bettors while exploiting their signals How Polymarket’s 0% fee model flips the game, removing friction for smart money and enabling a mandate to seek ground truth across politics, sports, and economics Why prediction markets achieve price efficiency but not signal efficiency—using the 2024 Trump-Kamala election as a real-world example of biased consensus The critical distinction between spot-market price convergence and Subnet 41’s meta-market that tracks signal consistency over time and volume How market efficiency correlates with dataset size—FX markets (massive, high-frequency data) vs. sports/politics (tiny, fragmented datasets with huge edge potential) Subnet 41’s store-of-value tokenomics: emissions capped by fees, 100% TWAP buyback, and closed-loop rebates that reward sustained +EV signal The 1% opt-in fee via the Sportstensor dApp, miner-placed bets on Polymarket, and how top performers can earn 10–15% of emissions Signal value to Polymarket: risk management, liquidity provision, and training sharper lines—even if indifferent to individual market outcomes Total addressable market math: capturing just $1B in annual volume = $10M in fees, achievable with 3 high-scale desks betting $1M+/day Data strategy: proprietary eSports feeds (real-time gold, kills, tower health), computer vision APIs, and partnerships (SportRadar, game servers) to level the field for “little guy” miners Convex vs. non-convex optimization in incentive design—why convex guarantees one global solution and prevents emission leakage or gaming Stephen’s advice for subnet builders: be ruthlessly principled, hard-code anti-dilution constraints, and treat your incentive mechanism like a bridge with guardrails Chapters 0:00 - Highlights 1:22 - Introduction 2:16 - Sportstensor (Subnet 41) 2:31 - State of Betting Markets 5:11 - Polymarket 0% Fee Impact 8:51 - Efficient Market Hypothesis? 11:50 - Price vs Signal Efficiency 14:41 - Market Size & Efficiency 17:27 - Meta-Market on Polymarket 18:46 - Signal as Store of Value 21:22 - Emissions & Budget Caps 24:01 - Who Places the Bets? 26:18 - 1% Fee Incentives 28:36 - Value of Signal Data 30:22 - Outsized Rewards Explained 32:24 - Why Cross-Market Winners 35:16 - Who Needs Risk Management? 37:20 - Market Maker Future? 38:46 - Total Addressable Market 42:15 - Training Toward Truth 43:28 - Data Limits in Sports 46:19 - Providing Data to Miners 49:22 - eSports Data Advantage 53:34 - Convex vs Non-Convex IM 57:29 - Emissions Distributions 1:02:53 - Subnet IM Design Advice