Additionally, a resource that's deleted from the cluster directly, not with Fleet, is not immediately re-created with Fleet. I expected Fleet to correct drift.

These were operator error and misunderstanding. Deleting a bundle does cause the cluster resource to be destroyed, and changes to bundles are correctly deployed. My confusion hinged on getting to a first successful deployment. I thought my changes were not propagating because I didn't know how to evaluate the health of the pipeline. The key was for me to observe events on the GitRepo resource. Then I could see and troubleshoot the errors blocking my changes.