forge-signal-discovery

Prevent lookahead bias in time-series feature engineering by applying horizon-aware feature construction and validation constraints. This Skill guides you through configuring features that respect temporal ordering, ensuring production safety and preventing future data leakage during model discovery.

Walk-forward cross-validation for time-series prediction is a purged, multi-stage validation method that respects temporal data ordering. This Skill implements it as part of a three-stage validation process to evaluate model performance sequentially across time periods before deployment.

Build a configuration-driven signal discovery pipeline by defining data sources, time resolution, and history to trigger the guided workflow. This Skill automates experiment configuration, artifact generation, and model documentation for production-ready deployment.

Yes, you can use this workflow for diverse data sources in time-series prediction tasks. The data-first model discovery process begins with data inventory and exploration to map available sources, time resolution, and historical depth before engineering features.

The best way to validate ML pipelines before production deployment is using a multi-stage validation framework with purged walk-forward cross-validation and a deployment gate. This Skill enforces evaluation metrics and model documentation to ensure deployment readiness.

End-to-end signal discovery requires data science or ML engineering experience to navigate horizon-aware feature engineering and multi-stage validation. This Skill provides a structured, configuration-driven workflow but assumes familiarity with time-series prediction tasks.