Problem
Register and matter data must enter production without stopping desks that still have clients calling.
What happened
Pressure toward one elegant cutover. Failure modes hide until everything is live — then everything is on fire.
Why it happened
Diagram elegance and calendar optimism beat operational humility.
What changed
Migrate in slices aligned to workflows. Expand-contract schema changes. Keep the desk moving.
What I’d do differently
Schedule slices by desk risk, not by ERD beauty. Instrument each slice before the next.