Strangler-fig migrations without the drama

Every legacy migration proposal arrives with a cutover date. Almost none of them survive contact with the actual system, because the cutover date assumes you understood the legacy behaviour completely before you started.

Route by slice, not by module

Pick a user-visible slice with a measurable business result. Route its traffic to the new path behind a flag, and keep the old path warm until the new one has proven itself under real load.

Every slice ships with a rollback

If a slice cannot be rolled back in under a minute, it is too large. Reduce the scope until it can.

Measure the business result, not the migration

"60% of endpoints migrated" is not progress a stakeholder can use. "Checkout latency down 340ms for 60% of traffic" is.