Platform · Infrastructure & Tooling
Synth
Bittensor Subnet 50. Decentralized ML competition producing probabilistic price forecasts (1,000 simulated paths per asset, per horizon) for crypto/equities/commodities. API has direct Polymarket and Limitless comparison endpoints.
- Mechanism
- Infrastructure & Tooling
- Chain
- Bittensor Subnet 50 (TAO). API is chain-agnostic. PM integrations with Polymarket (Polygon) and Limitless (Base).
- Status
- Live. v1.9.0 released May 22, 2026. ~200 active miners. 12 assets on 24h competition (BTC, ETH, SOL, XAU, SPYX, NVDAX, TSLAX, AAPLX, GOOGLX, XRP, HYPE, WTIOIL); 5 assets on 1h HFT (BTC, ETH, SOL, XAU, HYPE).
- Founded
- 2025
- Market types
- Probabilistic price forecasts (1h + 24h horizons across 12 assets)
- Audience
- Quant traders · Crypto-native · AI/Agents · Builder · DeFi-native
- Website
- www.synthdata.co
- @SynthdataCo
Pricing mechanism
CRPS (Continuous Ranked Probability Score) scoring of miner ensemble forecasts. Miner emissions allocated via softmax over rolling-window leaderboard. Two parallel competitions split 50/50 (24h with β=0.1 over 10 days; 1h HFT with β=0.2 over 3 days).
Settlement
Not a venue; no position settlement. API consumers query meta-leaderboard ensembles (default 6-day high-frequency, 14-day daily). Pyth Oracle recommended as ground-truth reference at start_time and at each increment.
Strengths
- Novel mechanism: full-distribution probabilistic forecasts driven by CRPS-scored decentralized competition, qualitatively above point-estimate oracles (Pyth/UMA/Chainlink)
- Real money on the line for forecasters ($2M+ in TAO emissions paid since Feb 2025); incentive-aligned competition vs reputation-only platforms like Metaculus
- Direct Polymarket and Limitless comparison endpoints make it the only PM-adjacent infrastructure layer that natively answers 'is this market mispriced vs model consensus'
- Multi-asset coverage across crypto, equities, commodities is unusual for a crypto-native data product
- Open source MIT with transparent public leaderboard
- Mode Network parent (James Ross) provides ecosystem distribution and credible team
- CRPS scoring is industry-validated for probabilistic forecasts (same primitive used in academic weather forecasting and Metaculus calibration literature)
Weaknesses
- Not a prediction market; included in PM Atlas via integration endpoints, not as a venue
- Bittensor exposure: subnet economics depend on TAO emissions, alpha-token dynamics, and SN50 emission share
- API monetization unclear (no public pricing page, customer roster undisclosed)
- Self-published trial numbers (110% return on $2k Polymarket account) need independent verification
- Asset coverage is 12 deep, no FX/bonds/longer-tail crypto
- Mode Network dependence for distribution and brand
- Competition risk from centralized closed-source model providers undercutting on price
About this mechanism
Aggregators, terminals, oracles, Telegram bots, smart-contract primitives. Partners more than competitors · earn from PM activity without taking the other side.