Systematic framework for reproducible continual anomaly detection benchmarks

This framework establishes principled methods for defining, filtering, and ordering tasks in tabular continual anomaly detection (CAD). It provides five benchmark-ready scenarios derived from cybersecurity datasets to ensure task boundaries and distributions are reproducible and scientifically valid.