Long Lian focused on stabilizing experiment telemetry in the menloresearch/verl-deepresearch repository by addressing a persistent issue with WandB run termination. Using Python and leveraging strong debugging and logging skills, Long implemented a destructor-based approach to ensure wandb.finish is reliably called when the tracker object is destroyed. This fix resolved a problem where incomplete termination signals caused runs to be mislabeled as crashed, improving data integrity and synchronization between the tracker lifecycle and WandB telemetry. The work demonstrated careful attention to the nuances of software lifecycle management, resulting in more reliable experiment tracking and reduced risk of missing or inaccurate data.