Health systems rarely lack a plan for interoperability. They lack a plan that survives contact with the clinical system they already run, which cannot be replaced on the timeline the mandate assumes.
The useful reframing is that a standard interface is a contract about how data is exposed. It says nothing about how the data is stored, and it does not require you to change that first.
A façade is not a compromise
Putting a standards-conformant API in front of an existing system is sometimes treated as an interim measure to apologise for. It is usually the correct architecture, and it stays correct.
The façade owns the mapping between your internal model and the external one. That mapping is where the genuinely hard work lives, and it does not become easier if you rewrite the storage underneath it first. It becomes harder, because you are now changing both sides at once.
Identity is the hard part
Resource mapping is tedious and tractable. Patient identity is neither. The same person exists several times across systems, with names spelled differently, one record merged in 2018, and a duplicate created during an outage when the master index was unreachable.
Every interoperability programme eventually discovers that its real dependency is a trustworthy identity service, and the ones that discover it early do better than the ones that discover it during integration testing.
- Decide what a match means before you write matching code
- Make merges reversible, because some of them will be wrong
- Keep an audit trail of every merge and unmerge
- Expose confidence, rather than silently picking a winner
Conformance is a floor
Passing a conformance suite means your responses are shaped correctly. It does not mean a receiving system can do anything useful with them.
The gap is usually in what is optional. Two conformant implementations can disagree about which fields are populated, which code systems are used, and what an absent value means. Those disagreements only surface when a real partner integrates, which is why the first integration should happen far earlier than feels comfortable.
Sequence by clinical value
Programmes that sequence by resource type spend a long time before anything is usable clinically. Programmes that sequence by workflow deliver value early and learn faster.
Pick one workflow that a clinician performs daily, make it work end to end including the unhappy paths, and put it in front of the people who will use it. What they say in the first week will reorder your backlog more usefully than another quarter of specification work.
