Reachable
Can a process read the records at all, on a schedule, without a person exporting a file? This is the least glamorous of the three and the one that most often blocks everything downstream for months.
The scope here is narrower than it looks. You need the fields the first use case consumes, not the warehouse. Teams that begin by modelling every table they own are still modelling a year later, and the use case that justified the work has been reassigned.
Joinable
Can a record in one system be tied to the same subject in another, reliably enough that a wrong join is rare and a missed join is measurable? Pharmacy, clinical records and claims each describe the same patient in their own vocabulary, and the quality of any analysis across them is capped by the quality of that link.
Measure the link before trusting anything built on it. A match rate is a number you can put in a report; "the systems are integrated" is not.
Governed
Who may see what, under which agreement, and what gets logged. This is the job that genuinely needs the committee, which is why it starts on day one and runs in parallel with the other two rather than waiting for them.
Running it in parallel is the whole trick. Governance discussions have their own clock, and starting them after the technical work finishes adds their full duration to the project.
Six weeks, sequenced
Weeks one and two establish reachability for the fields the first use case needs, and open the governance conversation. Weeks three and four build and measure the join. Weeks five and six close the governance position and hand over the pipeline with its match rates documented.
What that produces is a dataset one use case can be built on, with its limits written down. It is a smaller claim than "the data is ready", and it is one that survives contact with the next question.
