MiroFish Scenario Prediction
Overview
MiroFish is a swarm intelligence engine that builds simulated digital worlds from real-world seed data. Thousands of agents with distinct personalities, memory, and behavioral logic interact to produce qualitative scenario predictions.
Complementary to TimesFM: TimesFM gives quantitative price curves with confidence intervals. MiroFish gives qualitative scenario narratives — what happens if a specific event occurs and how it cascades through markets and public opinion.
When to Use
- "What happens if the Fed raises rates?" — scenario simulation
- "Impact of a whale dumping 10K BTC?" — cascading effect analysis
- Public opinion modeling for token launches or policy changes
- Complementing TimesFM forecasts with narrative context
- Any council question in the trading or strategy domain
- Strategic decisions that need "what if" scenario exploration
How to Start MiroFish
# Backend (Python 3.10, port 5001)
source ~/Projects/MiroFish/backend/.venv/bin/activate
cd ~/Projects/MiroFish && python backend/app/main.py
# Frontend (port 3002 — avoid conflict with Agent Orchestrator on 3000)
PORT=3002 npm --prefix ~/Projects/MiroFish/frontend start
Prerequisites: Real API keys in ~/Projects/MiroFish/.env:
LLM_API_KEY— OpenRouter or Claude API keyZEP_API_KEY— Zep Cloud key for agent memory
Integration with Council
When /council is invoked for trading or strategy domains:
- Phase 0: If MiroFish is running, seed a scenario simulation with the council question
- Feed the simulation results as additional intelligence to the advisors
- The Contrarian advisor should specifically challenge MiroFish's assumptions
- The Executor should reference both TimesFM numbers and MiroFish scenarios
Integration with /plan
When /plan routes a trading or strategy task:
- Use MiroFish for "what-if" analysis before committing to a trading strategy
- Compare MiroFish qualitative predictions against TimesFM quantitative forecasts
- If predictions diverge significantly, escalate to council decision
Cost Warning
Each simulation runs thousands of LLM-powered agents. A single scenario can consume significant API tokens. Use judiciously — prefer TimesFM for routine forecasting, reserve MiroFish for high-stakes scenario analysis.
Combo: Full Prediction Stack
Aster MCP (live market data)
→ TimesFM (quantitative: price curves + confidence intervals)
→ MiroFish (qualitative: scenario narratives + cascading effects)
→ Council (multi-advisor synthesis of both)
→ autoresearch (measure prediction accuracy over time)