Ouroboros – Distributed Self-Improvement¶
Status: TRL-4 (laboratory-validated)
Ouroboros is the recursive self-improvement loop of the BlackSwan swarm. It combines a genetic engine, champion/challenger deployment, and distributed genome exchange to continuously optimise trading strategies.
Key components¶
- GeneticEngine (
sim/genetic_engine.py) – population-based evolution with crossover, mutation, and elitism. - Champion/Challenger – new strategies are shadow-tested before replacing the active champion.
- Genome exchange – nodes share their best genomes via Redis pub/sub (legacy) or CRDT‑Gossip (decentralised).
Formal verification¶
The invariant V_s > V_h (improvement rate exceeds degradation rate) is
proved in formal/tla/Ouroboros.tla and monitored live in the swarm logs.
Observed performance¶
- Kelly parameters converge to
max_risk_per_trade ≈ 0.20, phi_llm ≈ 0.15. - Capital growth ratio exceeds 7× compared to static strategies.