One cell, five columns

Every spreadsheet eventually grows a column called Address with 12 Mill Lane, Cambridge, CB1 2AB, UK in it. It reads well. It is also the moment you lose the ability to answer "how many customers are in Cambridge", because the city is now a substring.

The database answer is five columns, and then the grid is unreadable — five narrow cells you scroll past to read one address. Display groups are the way out of that trade: one cell to read, five columns to query.

The group is a reading, not a column

Pick the columns, pick a separator, and they render as a single cell in position order. What matters is what happens underneath: the group never exists in SQL. The SELECT lists each member column individually and the join happens after the rows come back.

That one decision is why nothing is lost:

• Sorting a group orders by every member, in position order. Sort a Name group of first and last and you get first-then-last, which is what you wanted and what sorting a concatenated string would not have given you.
• Filtering happens on the columns, because the columns are still there and still typed. city = Cambridge is an indexable equality on a real column, not a LIKE against a sentence.
• Exports and the API see the parts. Nothing downstream inherits the presentation choice.

Each part keeps its own formatting

The members are formatted before they are joined, each with its own display options. A SELECT column contributes its label rather than its stored key. A boolean contributes Yes or No. A number keeps its thousands separator and its decimal places; a date keeps its format.

So a group of Status · Amount · Due Date reads as Overdue · 1,240.00 · 14 Mar 2026 and not as 2 · 1240 · 2026-03-14T00:00:00Z. Formatting you configured once on the column follows the value into the group.

Two smaller behaviours in the same spirit. Nulls are skipped, so a missing middle name does not leave you Jane Smith with a hole in it or a dangling separator. And hidden members are skipped, so you can keep a column in the group but out of the reading.

The group behaves like the sum of its parts

Its capabilities are derived rather than declared, and the direction of each default is deliberate:

Sortable is generous because one sortable member is enough to make the click useful. Read-only and hidden are strict because a group that hid an editable column, or hid a visible one, would be a group that lost you access to your own data.

Editing edits the parts

Opening a grouped cell gives you a form with one field per column, each with its own type, widget and validation. You are never handed the joined string to re-parse, which is the failure mode of the single-text-column approach — the point at which somebody's comma breaks your address.

What it doesn't do (yet)

• Assembled per row, per query. The joined value is computed after the rows return, so there is nothing to index and nothing to filter against as a whole. Filter the members.
• One separator per group. You cannot have a comma between the street and the city and a newline before the postcode. Real addresses want that; today you get one character or string, used everywhere in the group.
• A skipped null is silent. Because nulls collapse cleanly, a group is not the place to notice that a value is missing — it looks the same as a value that was never part of the group. If the presence of a field matters, keep it visible on its own.
• No nesting. A group cannot contain a group.
• Position order only. The reading order is the column order; there is no separate ordering for the group.