Auto-group

The lowest-friction path: hand the grid a flat frame and nothing else. With auto-group on, schema inference picks the row hierarchy — the non-numeric columns, ordered by ascending cardinality so the coarsest dimension nests outermost — and turns every numeric column into a summed measure. It is a starting point, not a straitjacket: any explicit rows / measures you pass always win.

Dash (Python)

from dash_tensor_grid import TensorGrid, build_grid

# No rows / measures — the engine infers both from the data.
payload = build_grid(sales, {
    "auto_group": True,
    "formats": {"revenue": "currency:USD", "cost": "currency:USD"},
})
TensorGrid(id="grid", row_data=payload["row_data"], column_defs=payload["column_defs"])

React

<TensorGrid
  data={data}
  autoGroup
  formats={{ revenue: '$0,0', cost: '$0,0' }}
/>;

Vue

<TensorGrid
  :data="data"
  auto-group
  :formats="{ revenue: '$0,0', cost: '$0,0' }"
/>

Vanilla JS

mountTensorGrid(el, {
  data,
  autoGroup: true,
  formats: { revenue: '$0,0', cost: '$0,0' },
});

Notes

• Dimensions by ascending cardinality. Columns with fewer distinct values group first, so a region → country → city-shaped frame nests in that order without you naming it.
• Numeric columns become sum measures by default. Pass an explicit measures map to change an aggregation or drop a numeric column from the roll-up — your map takes precedence over inference.
• Cap the depth. Dash accepts max_levels (engine.infer_schema) to limit how many dimension levels are derived; omit it for the engine default.
• Explicit always wins. Provide rows and auto-group only fills in the measures, or vice versa — inference runs only for the axis you left empty.
• Once you know the shape, prefer spelling out aggregation rows/measures for a stable, reviewable config.