How 11 Nightly Engines Evolve Trading Strategies While You Sleep

Most trading platforms give you a fixed set of strategies and call it a day. Strategy Arena takes a radically different approach: every night, 11 independent engines wake up, evaluate the entire strategy population, breed winners, mutate parameters, and retire underperformers. By morning, the arena is sharper than it was yesterday.

This is not a marketing gimmick. It is inspired by the same principles behind Andrej Karpathy's autoresearch concept -- autonomous systems that improve themselves without human intervention.

What Is the Evolution Lab?

The Evolution Lab is Strategy Arena's self-improving backbone. It runs 11 specialized engines in sequence every night, each with a distinct job:

1. Performance Scorer -- ranks all active strategies by risk-adjusted return across multiple timeframes.
2. Parameter Mutator -- takes top-performing strategies and slightly tweaks their parameters (lookback windows, thresholds, position sizes).
3. Crossover Breeder -- combines logic from two strong strategies into a hybrid offspring.
4. Decay Detector -- flags strategies whose edge has been fading over the last 7-14 days.
5. Regime Tagger -- labels current market conditions (trending, ranging, volatile) so regime-aware strategies can adapt.
6. Fitness Validator -- backtests mutated and bred strategies against recent data before promoting them.
7. Diversity Guardian -- ensures the population doesn't collapse into clones of one approach.
8. Risk Auditor -- checks that no evolved variant exceeds drawdown or volatility limits.
9. Archive Manager -- stores retired strategies and their historical DNA for future resurrection.
10. Changelog Writer -- logs every mutation, birth, and retirement into the Living Wiki.
11. Graph Updater -- refreshes the Knowledge Graph with new relationships between strategies, parameters, and market regimes.

Why 11 Engines Instead of 1?

A single optimization loop tends to overfit. It finds what worked yesterday and doubles down until it breaks. By splitting responsibilities across 11 engines with competing objectives -- performance vs. diversity, aggression vs. risk control -- the system maintains a healthy population of strategies that can handle different market conditions.

Think of it as checks and balances for algorithms.

What Actually Changes Overnight?

Each morning, you might see on your Dashboard:

• Parameter shifts: A momentum strategy's lookback window moved from 14 to 17 periods because the Regime Tagger detected a slower trend cycle.
• New hybrids: The Crossover Breeder combined elements of Turtle Trading and CUDA Evolved into a variant that outperformed both parents in validation.
• Retirements: A strategy that ranked bottom-5 for three consecutive weeks got archived. Its DNA is preserved -- if market conditions shift, it can be resurrected.
• Risk adjustments: The Risk Auditor reduced position sizing on a volatile strategy after drawdown crept above threshold.

All changes are logged with full transparency. Nothing happens in a black box.

How Does This Compare to Traditional Optimization?

Traditional backtesting optimization runs once, finds "optimal" parameters on historical data, and deploys them forever. The problem: markets change. Parameters that crushed it in a bull run get destroyed in a choppy range.

The Evolution Lab re-optimizes continuously. It does not chase yesterday's returns -- the Fitness Validator tests against rolling windows, and the Diversity Guardian prevents the entire population from converging on one regime's winners.

This is closer to how biological evolution works: constant pressure, constant adaptation, no finish line.

Can You See the Evolution History?

Yes. The Living Wiki documents every change, and the Knowledge Graph visualizes how strategies relate to each other -- which ones share parameter DNA, which ones were bred from common ancestors, and which market regimes favor which lineages.

Getting Started

Visit the Evolution Lab to see tonight's scheduled engines, recent mutations, and the current fitness leaderboard. No configuration needed -- evolution runs automatically for all strategies in the arena.

FAQ

How often do the 11 engines run?

Every night during low-volume hours (typically 2-4 AM UTC). The full cycle takes 15-30 minutes depending on population size. Results are visible on the Dashboard by morning.

Can evolution make a strategy worse?

Mutations are validated before deployment. The Fitness Validator backtests every change against recent data, and the Risk Auditor enforces drawdown limits. If a mutation fails validation, it is discarded -- never promoted to production.

Is this the same as Karpathy's autoresearch?

It shares the core principle: autonomous systems that evaluate and improve themselves without human prompts. The difference is domain -- autoresearch targets ML experiments, while the Evolution Lab targets trading strategy parameters, logic combinations, and risk controls.