# Understand the document model

Learn nodes, formatting and Schema, then choose a structure for articles or Word documents.

The previous chapter passed HTML to `content`. Core converts that input into a document tree describing the content; it does not store a page screenshot.

## How a paragraph becomes a document

A paragraph containing “Hello, QuireKit” has this JSON representation:

```json [One paragraph]
{
  "type": "doc",
  "content": [
    {
      "type": "paragraph",
      "content": [{ "type": "text", "text": "Hello, QuireKit" }]
    }
  ]
}
```

`doc`, `paragraph` and `text` are nodes. Bold, italic and links are marks attached through a text node's `marks`. Extra information such as heading level belongs in `attrs`.

Commands can modify this structure, history can undo changes, and different views can display the same content. Page numbers and selection pixels are results of layout and display.

## Schema defines allowed content

Schema is the set of structural rules: which headings exist, what a list may contain, and which marks text supports. Core builds it from the installed Node and Mark extensions.

StarterKit already supplies common structures and formatting. Add a Node when introducing a new content type, such as the [note group](/en/build/extension).

Restoring saved content requires the matching extensions. Unknown nodes must not silently become ordinary paragraphs or disappear.

## Choose articles or Word documents

| Product needs | Start with | Difference |
| --- | --- | --- |
| Notes, articles, knowledge bases | StarterKit and the generic document model | Headings, paragraphs and lists follow content order |
| A Word workspace like Classic | Word extension preset and Word document model | Also stores sections, headers, footers, paragraph styles and numbering |

A Word heading may be a styled paragraph rather than a `heading` node. Wrapping ordinary paragraphs in a page border is not a Word document model, and switching renderers does not add its content rules.

In `assembly/word.ts`, `createClassicWordPreset()` supplies compatible extensions and `createClassicWordDefaultDocument()` supplies matching content. This does not create the Classic toolbar. The preset handles overlapping StarterKit nodes, so do not append a complete StarterKit again.

## How the model relates to layout

The model describes the content. A document profile supplies its page, paragraph and table rules. View displays layout results and handles input.

Articles can use continuous layout; Word extensions supply section and pagination rules. Reuse the preset initially. The [profile and layout API](/en/api/extensions) is for implementing new models or layout capabilities.

## Read and save content

```ts [Read editor content]
const document = editor.state.doc
const selection = editor.state.pm.selection
const backup = editor.getJSON()
```

Read structure through `state.doc` and selection through `state.pm.selection`. Use `getJSON()` at save boundaries rather than maintaining a second reactive document and replacing content on every edit.

Save your model version too. The complete example distinguishes `generic-v1` and `word-v1`, validating version and Schema before restore. Switching Canvas/SVG changes presentation without changing the model.

Next: [display the document with Canvas/SVG](/en/build/view). Both complete assemblies are in the [workspace reference](/en/build/reference).