The darwintIQ Genetic Algorithm Engine
Overview
darwintIQ maintains a continuously evolving population of trading models that adapt to recent market behaviour.
Rather than optimising a fixed strategy offline, darwintIQ continuously generates, mutates, and evaluates models on a rolling window of live market data. This creates an adaptive model ecosystem whose fitness changes as market conditions shift.
Model Population
darwintIQ operates on a population of candidate trading models.
A Trading Model is consists of:
- Entry Logic (how the Trading Model enters trades)
- Position Manager (how it manages risk & exits)
- Regime Filter (when it is allowed to trade)
New variants are produced through mutation of existing models. Variants with poor fitness are discarded, while better-adapting variants persist.
This results in an evolving model population rather than a static strategy set.
Rolling Evaluation Window
All models are evaluated on a rolling four-hour window of recent market data. The window advances forward continuously in time:
- every minute, the oldest minute of data drops out
- the newest minute of market data enters
Models are therefore always evaluated on the most recent four hours of behaviour. darwintIQ does not evaluate models on fixed historical segments or static backtests.
darwintIQ continuously re-evaluates Trading Models on recent market data and ranks them by metrics such as fitness, entry quality, and entry-move context.
Continuous Mutation
Both strategy parameters and position-management parameters mutate over time.
Mutation occurs continuously as part of the population process. There is no discrete optimisation phase.
As a result:
- the population explores nearby behavioural variants
- adaptation occurs incrementally
- model structure remains comparable across time
Fitness Definition
Model fitness in darwintIQ is multi-dimensional. It is not determined by profit alone.
Fitness incorporates:
- profitability
- stability of returns
- consistency of trade outcomes
- distributional behaviour of trades
Metrics used in fitness estimation include:
- standard deviation of returns
- Sharpe ratio
- Jensen–Shannon divergence of trade distributions
- win/loss structure
- related behavioural statistics
Fitness therefore reflects how stable and consistent a model’s behaviour is under recent conditions, not merely how profitable it was.
Relative Evaluation
Models are evaluated relative to other models in the population.
Fitness is therefore contextual rather than absolute.
This allows users to see:
- which models are currently adapting best
- which models are deteriorating
- how behaviour shifts across the population
darwintIQ emphasises comparative fitness rather than standalone performance numbers.
Temporal Fitness Dynamics
Because evaluation occurs on a rolling window, fitness changes naturally as market behaviour shifts.
darwintIQ tracks how model fitness evolves through time.
This reveals:
- adaptation
- deterioration
- regime sensitivity
- behavioural instability
Fitness changes are inferred from observed performance dynamics, not from hypothetical tests.
What darwintIQ Does Not Do
darwintIQ does not:
- simulate hypothetical market scenarios
- perform stress testing
- run Monte-Carlo perturbations
- test parameter shocks
- evaluate models on synthetic regimes
All evaluation is based on observed behaviour in recent real market data.
Interpretation Model
darwintIQ is an observational adaptive system.
It observes how models behave under changing conditions and tracks their relative fitness across time.
Users interpret model fitness together with:
- population comparison
- temporal change
- behavioural consistency
rather than relying on a single performance metric.
Conceptual Positioning
darwintIQ can be understood as:
- an online evolutionary trading system
- a rolling adaptive model population
- a live fitness estimation framework
rather than a traditional backtesting or optimisation platform.