how-to-schema-and-warehouselisted

# how-to-schema-and-warehouse Infrastructure-layer skill (the base of the phoebe-data-skills 4-layer roadmap). Trust-first: the output is a warehouse someone can actually query plus a contract that keeps it trustworthy - never a pretty ERD of already-clean data. Showcase walkthrough (Everrest retail case, real DuckDB build): https://github.com/phoebefu6/phoebe-data-skills - `docs/how-to-schema-and-warehouse/` ## The lineage this skill produces `raw dump (messy sources) -> lake (partitioned) -> warehouse (star schema) -> marts`. Defects live in the raw layer; the warehouse transform is where they get resolved. ## The 6 steps ### 1. Input Collect the raw sources as-is: every export, its key format, its grain, its defects. Note where keys should join across systems. Do not assume anything is clean - a real platform is handed exports, not tables. ### 2. Sample data (only when real data isn't available yet) Write a seeded generator producing the raw dump from multiple "source systems" with ingestion-flavored defects planted on purpose: orphan foreign keys, heterogeneous keys across systems, dirty categorical labels, timezone/timestamp bugs, fabricated amounts, duplicate identities. Document each in the docstring. With real data, skip - the defects are already there. ### 3. Objective Frame the trust question: "can we turn this dump into a warehouse the company can query and trust, and what breaks if we skip modeling?" 3-5 sub-questions covering keys resolving, sources reconc