# Continuous Model Refinement

* **Nightly Retraining Pipeline**\
  – Ingests the day’s new price, on-chain and sentiment data to retrain models every 24 hours.\
  – Applies online learning techniques that prioritize recent data while retaining long-term patterns.
* **Performance Feedback Loop**\
  – Compares predicted fair-value bands and sentiment calls against actual market outcomes to identify model drift or overfitting.\
  – Adjusts feature weights and hyperparameters automatically to improve next-day predictions.
* **A/B Strategy Testing**\
  – Runs parallel simulations of updated models against historical data to validate improvements before deploying to live accounts.\
  – Only promotes model versions that demonstrate higher expected return and lower drawdown risk.
* **Governance and Oversight**\
  – Maintains version control on model code and datasets for full transparency.\
  – Conducts regular reviews by our quant research team to ensure models remain robust against emerging market conditions.