Strategy Evolution

Strategy Evolution is a research engine that automates the entire quant workflow:

1. Generate — each cycle, ~20 new strategy genomes are spawned via crossover + mutation from the top-quartile survivors. Population diversity is enforced (no monoculture). 2. Backtest — every candidate runs through walk-forward OOS validation. 3. Stat-gate — three independent tests must pass: DSR (Deflated Sharpe Ratio), White's Reality Check (against the full search universe to avoid survivor bias), and PBO (Probability of Backtest Overfitting). 4. Paper-trade — survivors run live in paper mode against real market data. 5. Lifecycle — Qualified → Paper → Canary → Live. Auto-kill on Sharpe degradation or DD breach.

Public dashboard: `/lab` → Evolution tab. Admin controls: force promote, kill switch, manual reseed.

Evolution is additive, not destructive:

• The locked Phase 37 baseline keeps trading regardless of what Evolution discovers
• Candidates ship at weight = 0 and stay in shadow until 28+ days of OOS validation
• Promotion to live requires admin approval + 4-week green-window
• Every promotion, kill, and parameter change is logged to immutable lifecycle transitions
• The stat-gates run on the FULL candidate population — survivor bias and overfitting can't sneak through

Result: an always-running R&D pipeline that can adapt to regime shifts without ever putting your capital at risk during discovery.

Surviving strategies don't immediately enter the production allocator. They enter the 9-signal orchestrator ensemble as a conditional 10th signal, also at weight 0, and walk the same shadow-validation gauntlet as every other new signal (see `/help/ai-evolution`).

This is the unique part: Evolution doesn't just generate strategies — it generates calibrated signals that the orchestrator can blend with the existing 9. The system improves itself, but only through the same rigour we'd apply to an external researcher.