TailBooster introduces a dual‑layer generative framework that augments rare extreme events in air transport data while enforcing operational validity, ensuring synthetic instances are feasible for real‑world operations. By focusing on the distributional tails that are under‑represented in historical records, it provides more reliable training signals for machine learning models. The approach aims to mitigate cascading network disruptions, economic loss, and safety risks associated with severe arrival delays and abnormal flight times.