Skip to main content
Most columns hold one value. Some don’t: an address is a set of named fields, a contact has several phone numbers each with a type, an invoice has line items. Declaring that shape on the column means Vern validates it, instead of treating the cell as an opaque blob of text.

​
Declare the shape

Open the column in Templates and look for Each cell holds. Four choices: For a record or a list of records, add the Fields underneath. Each field has:
  • a name,
  • a type — Text, Number, Whole number, or Yes / no,
  • a Required toggle,
  • optional allowed values, entered comma-separated. Supplying them means the field accepts nothing else, which is how you pin an enum like mobile, home, work.
Clear the shape removes the declaration and the column goes back to holding one plain value.
Field-level Required is not the same as the column’s Required checkbox. The column checkbox asks whether the cell has a value at all; a field’s toggle asks whether a record that is present must carry that field.
A record with no fields declared accepts anything. That’s a useful escape hatch for “this holds an object and I don’t want to pin it down yet”, but it validates nothing.

​
How a structured cell is stored

The cell holds JSON. A list of records looks like this:
The import agent writes this shape, and the grid reads it. Validation checks the JSON against the declared shape: a cell whose contents can’t be parsed, or whose fields don’t match, is flagged like any other invalid cell.

​
The raw JSON Schema editor

Schema Structure appears underneath once a column already has a shape. It is the full JSON Schema behind what the visual editor shows, and it is the escape hatch for anything the four choices above can’t express:
  • nesting deeper than a record inside a list,
  • a non-text value at the top level — the visual editor’s A single value and A list of values both mean text, so a bare list of numbers is authored here,
  • length bounds (minLength, maxLength), numeric ranges, format,
  • minItems / maxItems on a list,
  • additionalProperties: false to close a record to exactly its declared fields,
  • unions and conditionals.
Two guarantees worth relying on:
  • The visual editor never rewrites your JSON. Opening a column and saving with no changes writes back byte-identical schema. Anything the editor doesn’t model it leaves alone — you’ll see “This column uses a custom shape. Edit it as JSON below” rather than a simplified version of what you wrote.
  • A broken schema is refused at save time, with the path to the problem — Column "Phones": contacts.number: "text" is not a JSON type. It is never stored in a state where it silently validates nothing.

​
Shape and pattern together

A column can carry both a Strict pattern and a schema. When it does, the schema wins — it is what gets checked, and the pattern is ignored unless the schema itself contains one. Put the pattern inside the schema — a "pattern" keyword on the relevant field — rather than relying on both. Shape checking also runs regardless of what the Validation Rule dropdown says. A column whose rule is None but which carries a schema is still validated against it.

​
Referencing other templates from inside a structured value

When the thing that has to exist in another template is a field of a record rather than the whole cell, say so with a reference path — one link rule covers both cases:

Link columns

Pointing a link rule at a field inside the cell.

​
Next