Revenue Search: Inside Bittensor Subnet Session with Marc from Numinous: Subnet 6
21 snips
Feb 9, 2026 Marc (Numinous), founder of a forecasting subnet building agent-based leaderboards and Eversight, a research API for traders. He discusses LLM forecasting focused on geopolitics, sports and macro. Topics include how miners (agent code) produce forecasts, leaderboards and rewards, anti-copy measures, scaling to many events, and the Eversight product and API launch.
AI Snips
Chapters
Transcript
Episode notes
Protect Winners But Scale Events
- Hide winning miner code briefly (three days) to limit immediate copying and encourage innovation.
- Scale event volume and categories to reduce volatility and copying advantages over time.
Scale Events To Reduce Noise
- Numinous targets rapid event scale: already 200 events/day and aiming for 1,000 then 100,000 to stabilize skill evaluation.
- More events reduce evaluation noise and make leaderboard results robust.
Miners Already Beat Gemini In Benchmarks
- Marc shared benchmark work showing Numinous miners beating Gemini on forecasting tests.
- He reported a growing margin of improvement and confidence intervals favouring Numinous.