The data model is data#

Decision#

Content types are not code. A Data Class (its fields, field types, relations, locales, layout) is a record that an administrator edits in the dashboard at runtime, and a Data Object is a record of a class with its attributes stored as structured data (JSON in PostgreSQL). Adding a field to "Product" is a change in the dashboard, not a release. The model follows Pimcore's concepts (classes, objects, relations, field collections), and a Pimcore installation can be imported.

Why#

• The people who know the business change it. A new attribute, a new content type, a different list layout: no developer ticket, no deployment, no migration.
• One model for many uses. Products, events, team members, form submissions and data shown on pages are all Data Objects, so permissions, publishing, languages, search, export and the API work the same way for all of them.
• Validation still exists. The attributes are stored flexibly but checked against the class definition (a schema is compiled from it) at the boundary where they are written, so "flexible" does not mean "unchecked".
• A proven shape. Pimcore-style classes are familiar to PIM/MDM teams, which makes import and mental mapping straightforward.

What it costs#

• No relational integrity inside attributes. Because values are JSON, the database cannot enforce a foreign key between objects; relations are resolved and checked by the application.
• Filtering is done in the application, over the rows of a class, not with generated SQL on JSON paths. This is simple and predictable, and fine for catalogues of thousands of objects; it is not tuned for millions.
• Changing a field type of a class with data needs care; the dashboard refuses some changes and reports others.
• One row per language. Each locale of an object is a full row (linked by a translation group), which duplicates non-localised fields.

When to revisit#

When a class needs to hold very large volumes with ad-hoc querying: the likely answer is indexed projections for chosen fields, not a different model.

Where it shows in the product#

Data pillar, data model, data and storage; plans advanced-data-modeling-plan.md.