Releasing a multi-portal platform without losing state between components

You can have a platform where every screen passes its tests and the complete workflow still falls over. An applicant submits a record in one portal, an operator sees an older version of it in another, and an external service confirms an action that neither portal shows. Look at any one of those components on its own and it seems fine. What's wrong is that they each have a different idea of the shared state.

The situations below are a composite of patterns from late-stage platform work. They're not one organisation and I'm not reconstructing a private project. I use the same approach for each of them: name the state, follow how it changes, find the smallest contract that's failing, and give the implementation agent and the human reviewer enough evidence to check the repair.