April 2026
Phase 0: Building the data pipeline that bootstraps the colony.
Phase 0 is human-driven research and live trading — not a placeholder before the “real work” starts. Every trade rationale, every killed hypothesis, every market regime we navigate feeds directly into the colony's initial knowledge. The autonomous system doesn't start from zero. It starts from accumulated intuition.
1. Shipped
Polymarket BTC prediction model
Built XGBoost model on 70k+ hourly samples (2017–2025). Tested 17 temporal pocket candidates. Found the edge is diffuse across time rather than concentrated in exploitable windows. Pivoted to a full-timeframe betting strategy.
2. In development
QQQ swing trading system
Production-grade algo built on Nautilus Trader (Rust/Cython core). ML pipeline with XGBoost entry confirmation, walk-forward expanding window training. Live execution via IBKR. Hosted on AWS.
3. Ongoing
Competitive landscape analysis
Deep source-level analysis of a dozen AI agent systems — including MiroFish, Roo Code, and Claude Code. Extracted adoptable architecture patterns and catalogued anti-patterns to avoid. Findings directly shaped the colony architecture design.
4. Active
Knowledge base
Every research finding, decision rationale, and strategy hypothesis tracked in a structured knowledge base. This becomes training data for agent initialization.
5. Scheduled June 2026
Trader interview pipeline
Structured interviews with professional traders across styles (trend-following, mean-reversion, macro discretionary) to extract decision frameworks for AI agent seed personalities.