Script utility per validare OOS strategie discovered durante run Phase 2.5.
Carica un JSON strategia (formato Hypothesis output), fetcha OHLCV via
Cerbero, esegue BacktestEngine + FalsificationReport + AdversarialReport,
stampa metriche annualizzate (CAGR, Sharpe, max DD, Calmar).
Esempio:
uv run python scripts/backtest_strategy.py /tmp/strategy.json \
--start 2018-09-01 --end 2026-01-01 --label my-strategy
Validato sui top 2 genomi Phase 2.5 (flat-ablation e fitness-v2-combo):
flat-ablation top overfit su 7y (-37%), fitness-v2 top regge (+143% in 7y,
CAGR +12.8%). Conferma che strategie con time gate temporal feature
generalizzano meglio di strategie con SMA crossover hard-tied al regime
del training period.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
