Paper first, capital second
No Cortana layer trades real money until it has demonstrated statistical edge on out-of-sample data and survived paper trading across multiple market regimes.
The principle
Validate ruthlessly. Walk-forward validation and purged cross-validation per López de Prado. No in-sample backtests presented as evidence of edge. An in-sample backtest is a story, not evidence.
Survive first, compound second
Position sizing, daily loss limits, and drawdown controls sit above every strategy decision. A 55% edge means nothing if leverage takes me out in a drawdown.
What “multiple regimes” means
A layer must survive paper trading across distinct market conditions - trend days, chop days, low-vol compression, expansion, news-driven dislocation. A strategy that only works in one regime is not validated; it is lucky.
Paper→live threshold
[No data yet] - specific validation metrics that graduate a layer from paper to a small live sleeve. Needs to be defined and committed to before Layer 1 takes real capital. The threshold must be set in advance so the decision is not negotiated under the pressure of a winning streak.
Tooling
- IBKR Pro + IB Gateway for paper and live execution
- Python + ib_async
- XGBoost (LightGBM backup) for tabular financial signals
- Walk-forward + purged CV already in place for Layer 1
See Also
projects/cortana.mdconcepts/three-layer-thesis.md- each layer must pass this bar independently