Mutator Docs

1. Install the CLI

pip install eas-mutator
mutator auth login <your-api-key>

2. Define a search space

# search/momentum_v1.yaml
name: momentum_v1
universe: us_equities_top_500
generators:
  - V0  # random
  - V1  # bayesian
  - V2  # genetic
features:
  - rsi_14
  - ema_20
  - ema_50
  - vol_30
  - regime_indicator
horizons: [5min, 1h, 1d]
validation:
  walk_forward_splits: 5
  is_oos_ratio_min: 0.5
  deflated_sharpe_min: 0.8
  psr_min: 0.95
  backtest_overfit_prob_max: 0.05
deploy:
  paper_first: true
  paper_days_required: 14
  drift_sigma_max: 2.0
  capital_ramp: [0.01, 0.05, 0.10, 0.25, 0.50, 1.00]
  kelly_fraction_max: 0.25

3. Run the search

mutator search start search/momentum_v1.yaml
mutator search status   # poll candidates surviving each gate
mutator promote <candidate-id> --to=paper

Core concepts

• V0-V6 hierarchy — 7 strategy generators of increasing sophistication. Candidates compete; only top survivors advance.
• Deflated-Sharpe — López de Prado's correction for multiple-testing bias in strategy search. Mutator never reports raw Sharpe.
• Brain-lock — prevents concurrent strategy edits from racing each other into live deploy.
• Closed-loop drift — paper PnL is continuously compared to backtest expectation. Drift > 2σ quarantines the strategy.
• Capital ramp — live strategies start at 1% of allocated capital and ramp up only after passing each tier's PnL/drawdown check.