School-lab containers inherit the wrong default runtime after a partial toolkit install. What should ops repair?

School-lab containers inheriting the wrong default runtime after a partial NVIDIA Container Toolkit install require completing the toolkit installation and correcting Docker runtime defaults on the civic lab hosts. A partial install can leave Docker configured for runc while NVIDIA runtime packages or daemon.json default-runtime settings are incomplete, preventing student workloads from entering the GPU path. Completing the toolkit setup and setting the intended default restores lab GPU access. Leaving the partial install and teaching students to ignore GPU errors normalizes a broken platform and wastes instructional time. Switching every lab to Windows containers avoids neither the GPU AI lab requirements nor the NVIDIA software stack targeted by the curriculum. Deleting Slurm does not repair Docker runtime defaults; many municipal education sites use Slurm or Kubernetes alongside Docker, and removing the scheduler also removes fair-share classroom scheduling. Exam caveat: distinguish the Docker default runtime from per-container runtime flags; fixing defaults helps interactive demos, while batch systems may still need explicit runtime annotations. Also distinguish toolkit completeness from driver installation: both are required, though the question identifies a partial toolkit and wrong default runtime. Practical check: rerun toolkit configuration with nvidia-ctk or site playbooks, verify daemon.json runtime entries, restart Docker, confirm docker info lists nvidia, launch a GPU class canary, and bake the completed configuration into the school-lab golden image to prevent regression on rebuild.