Recently, we've noticed that the newly built Jupyterlab images (after December 2025) do not expand the environment variables that are passed as pod arguments.
We rely on this feature to create generic pod defaults in Kubeflow, such that we place environment variables as placeholders in the pod arguments, and then the value of environment variable may be different for different namespace / profiles.
We need to investigate on this if this is a regression in the way we build our image, or a regression in pebble or rockcraft and fix it.