Make the model · 6
A model from sentences, checked by the business
The answers to “who has time to build this?” and “who is going to maintain this?” An analyst types what the business says and the model changes. The whole model then reads back as plain sentences, so the person who knows the business can say which ones are wrong, and fix them where they read them.
Saying it
- An analyst types what the business says. Prose, as it is spoken: “suppliers ship items in purchase orders.” Nobody draws anything.
- The language model answers with changes to the model. It is sent the model as it stands and the sentence, and returns a list of changes: things, the facts between them, and how many of each. It can add, correct, rename and remove.
- The changes are shown before they are made. Unless someone switches that off, they arrive as a draft over the drawing, with what the model now says and what it does not yet say, and they happen when someone commits them.
- One piece of code applies them. Every change, whoever proposed it, becomes part of the model the same way, and a change that cannot be right, a thing renamed onto another thing, is refused there.
Reading it back
- Whoever knows the business opens Read. The model as plain sentences, one page for each thing: “An order can be shipped by more than one shipper, or none at all.” No diagram to learn and no SQL to follow.
- They fix a sentence where they read it. Tap how many, and the alternatives arrive as finished sentences to choose from. A table with a cryptic name is given the business's word in place, while the SQL keeps the name the database knows.
- The correction is the model. It is saved like any other change and kept in the model's history, with nothing to write up and nobody to hand it to.
When they cannot tell. From any sentence a reader can also open it on the drawing, say it is wrong and let the analyst correct it, or send it to whoever answers for that table, which is scenario 8.