Worked on the kubernetes/autoscaler repository to deliver a targeted bug fix that improved atomic scale-down correctness in large-scale Kubernetes clusters. Addressed an edge case where the number of scale-down candidates matched the number of registered nodes, ensuring the autoscaler filtered unremovable nodes using only registered nodes. This approach prevented indefinite blocking of scale-down operations caused by failed instances, reducing latency and enhancing cluster efficiency. The solution was implemented in Go and Shell, leveraging expertise in autoscaling, cloud computing, and system design. The work strengthened production reliability and predictability for Kubernetes environments, focusing on operational impact and robust engineering practices.