A case study shows one program. The toolkit is what carries forward: the artifact library collected in one place, and the principles that make the difference between running a migration and getting run by one.
Artifact library
The nine documents that carry a migration. Each links to where it lives in the walkthrough.
The only-ness
Strip away the specifics and these are the reusable truths — the accumulated craft that turns a case study into a way of working.
The source-to-target mapping is the only place the two schemas and the business meaning agree. Own it and you own the risk; leave it ambiguous and it gets resolved at the keyboard, silently.
The cheapest scope you'll ever cut is the dead reports you refuse to carry forward. Retiring 38% of reports paid for itself before the build began.
Most reconciliation failures trace to a grain nobody stated out loud. Write it on both sides of every mapping row before anyone writes a join.
Snowflake won't enforce your keys. Integrity moves out of the schema and into the test suite — "migrated" isn't done, "reconciled to the penny" is done.
Running both platforms in parallel and reconciling daily means cutover removes the old system instead of gambling on the new one.
Adoption by thousands isn't a training week at the end. Awareness, involvement, training, adoption — threaded through every phase — is what senior TPMs do that coordinators don't.
Ambiguous fields get ruled by the business, logged, and never re-litigated. A decision log is cheaper than the same argument three times.
"80% done" means nothing; "every object in this wave reconciled and signed off" means the wave can exit. Gates make progress honest.
Behind the walkthrough
As the Technical Program Manager, I ran this cross-functional EDW-to-cloud migration end to end — from charter through decommission, not a single workstream inside it. The governance rhythm, the phase gates, and the judgment calls all reflect that ownership.
Every decision, trade-off, and artifact on these pages is one I can walk through in depth — the real craft of running large, cross-functional data programs.