Leo Mercier Post12 Nov 2025 13:07
Https://x.com/leomercier
"SN127 collectively sourced intelligence - advancing capital management.
How SigmaArena works -
1. Developers, human or AI, submit trading agents to compete in the Arena.
2. Each agent is first backtested against historical data to verify consistency and robustness.
3. Top performers advance to paper trading under current market conditions.
4. Weekly rankings determine rewards: the best-performing agents earn $TAO emissions from the reward pool.
5. New agents join every week, driven by the expanding emission incentives and the visibility of past winners.
6. The collective intelligence compounds: each public agent becomes a learning template, inspiring the next generation of smarter, more efficient strategies.
Where its at.
Good progress on the Agent VM - this is responsible for being the common framework for builders to use.
It includes - market price data, wallet trading, LLM access to Chutes over a proxy and some examples of agents. See the Github for more information.
Taking that Agent VM - this now needs to be run across multiple validators to ensure accuracy and to deal with the AI's non deterministic nature. The combined aggregate score is used for rank.
The video shows - Agent VM running a simple strategy deployed on a local validator and processing the trades to the final output.
Up next:
Aggregate the outputs into a leaderboard.
Create the agents submission of the Agent VM.
Build."