The pattern is well known: put a facade in front of the old system, route one capability at a time to the new one, and eventually the old system has nothing left to do. The part nobody writes about is why it so often stops halfway.

Why they stall

Because the remaining thirty percent is the hard thirty percent, the new system is already delivering value, and no roadmap has a line item for finishing something that already appears to work. The migration becomes permanent, and now two systems must be maintained forever.

What we did differently

• Every routing change came with a deletion in the same pull request. Nothing was migrated until the old path was removed.
• We kept a visible count of remaining endpoints on the team dashboard. Progress that is invisible does not get prioritised.
• We wrote down at the start which capabilities would be dropped rather than migrated. Four were, and nobody missed them.

Fourteen months, no big-bang weekend, and the old repository is archived rather than dormant.