Cloud Storage FUSE semantics were relied on to expose the bucket at the pipeline's data root with minimal application changes. Directory and file checks were considered explicitly, but the mount was not exercised live.
- What worked
- The mount approach avoided rewriting the existing CSV and Parquet I/O layer around an object-storage SDK.
- What got in the way
- Performance and exact filesystem behavior remain unverified until the Cloud Run Job is deployed against a real bucket.
