Spreadsheets

Models live in real spreadsheets: Excel-compatible workbooks whose tables are connected to datasets in the workspace. Refresh, and the data updates while the formulas hold.

What a model is made of

• A connection: a sheet range that is a copy of a dataset (often a transformation's output — the model's feed).
• Calculated columns on the connection: the variables and the result, as plain Excel formulas (=[@seats]*[@price_per_seat]). Where one segment's economics differ, a variant gives those rows their own formula.
• Checks: what must be true — a total that matches the finest grain, a share that sums to one. A failed check is loud.
• Hypotheses for the forecast: named claims with a growth per period, accumulating in the scenarios they are active in. History never differs by scenario.

The spreadsheet page

The workbook renders in the app with its formulas; the formula bar highlights what a cell reads. The About rail lists the connections, the calculated columns and the checks. Refresh copies the latest data from the connected datasets and recalculates. History lists the versions; any version can be viewed and made current again. New chart over a range starts a chart on the model's table.

Models are built and edited by agents, which know the modeling practice the platform teaches. Open the workbook in Excel when you want it there: the formulas are as written.

Faster with an agent — "build a bottom-up model of <market> in a spreadsheet: drivers from the datasets, one formula per segment where the economics differ, and a check on the total." The agent builds the feed, connects it, writes the columns and the checks, and charts it.