Migration planning starts with dependencies and constraints. Rehosting moves a workload with limited redesign; replatforming changes selected components; refactoring changes the application more deeply. The right choice depends on time, risk, compatibility, and operating requirements. A fashionable service isn't enough reason to rewrite a working dependency during a rushed migration.
A data lake commonly separates raw, validated, and curated objects. Prefixes, formats, catalog metadata, permissions, and lifecycle rules make that separation usable. Storing files in S3 doesn't automatically produce consistent schemas or trustworthy data. Keep raw provenance, handle late arrivals, and define who can publish curated outputs. In the edge task, duplicated source files must not silently become duplicated business records simply because every upload succeeded.