That file is the intent model. It lists what the product stores (fields), what it allows and forbids (rules), the paths people take through it (journeys), and what nobody has decided yet (open questions).
Start with the requirements you have and the part you will build next. It does not need to be complete. The team agrees on it, then designers and engineers build from it. Small checks compare the build to the model. When a requirement changes, you change the model first and the screens after.
Handle changes at the source
Some changes only move things around on one screen. Others change a rule that several screens rely on. Ask one question: does anyone else depend on this? If yes, it goes in the model. If no, just change the screen.
5 views of the workflow
The same process, drawn 5 ways. Use whichever one makes sense to the person you are explaining it to.
The contract
Product managers and clientsHow to use it
Go through the model together. Everyone can see what is agreed and what is still open.
What to account for
It only helps if someone keeps it up to date.
The 3 lanes
Team leadsHow to use it
Each team knows what it receives and what it hands on. The handoffs are files, not meetings.
What to account for
A small team may have one person covering several lanes.
The change loop
Designers and engineersHow to use it
Write the request down, agree on it, then rebuild and check.
What to account for
Rebuilding takes time. Before changing a shared rule, see which screens use it.
The layers
Designers and frontend engineersHow to use it
How a screen looks comes from the design system. What is on it and how it behaves comes from the model.
What to account for
Spacing, density, and visual balance still need judgement on the screen.
The shared plan
The whole teamHow to use it
Everyone works from the same plan, and it gets updated as the team learns.
What to account for
The model grows during the build. It does not need to be finished on day 1.
What still needs review
The checks can tell you the build matches the model. They cannot tell you the model is right, the design is good, or the parts outside the screens work. People still review those.
Some checks reach a little further. Real data validation tests the app's data shapes against real responses saved from the server. For rules no script can check, write down that a person must look, so nobody assumes it is covered.
Watch every check fail once
A check that always passes might be working, or it might be broken. From the outside you cannot tell. Think of a smoke alarm nobody has ever tested: silence could mean no fire, or a cut wire.
So every check here comes with a small test file that breaks the rule on purpose, called a probe, and one that follows it. Run the check on the probe and watch it fail. Now you know it can. We call this the red gate. And if a check could not run at all, that is not a pass. It is a check that did not happen.
Set up your project