Pivot builder

EnterorSpace on a field

Opens the field menu: add a dimension to rows, columns or filters, or add a measure to values with a chosen aggregation.

EnterorSpace on a chip

Opens the chip menu: aggregation and heat shading for values, shown values for filters, move, reorder and remove.

Alt+Arrow leftorArrow right

Moves the focused chip one place within its well.

DeleteorBackspace

Removes the focused chip from the pivot.

Arrow keys in a menu, Escape

Move through menu items and close the menu, returning focus to the chip.

EnterorSpace on a row group

Expands or collapses that group. The corner button expands or collapses all groups.

• The result is a table with role="table", row, rowheader, columnheader and cell roles, and aria-colspan on grouped column headers, so screen readers read each value with its row and column.
• Every drag has a menu path: fields and chips are real buttons with menus, so the whole pivot can be built without a pointer.
• Row groups are buttons with aria-expanded. The scroll area is focusable and named, so keyboard users can scroll wide pivots.
• Changes are announced in a polite live region, such as "Country moved to columns" or "Exported ridgeline-sales-pivot.csv".
• Heat shading is never the only signal: every shaded cell prints its value, and the footer names the shaded measure with a low to high key.
• Pointer focus draws no ring; keyboard focus shows Arc's shared focus ring. Keyboard position shows as a filled chip, field or row header.

• The dragged chip follows the pointer on motion values, with no React render per frame. React only hears when the target well or insert position changes.
• Chips in the wells glide aside to open a dashed slot where the chip will land, on the smooth spring with no overshoot. On release the chip flies into that slot, or home if nothing changed, or fades out when dropped outside to remove.
• The table previews the drop while you hover: a new structure crossfades in, rows that appear or leave slide the rest into place, and a changed number rises into its cell.
• Group chevrons rotate on the spring token, heat shading fades by opacity, and the swap icon turns half a turn per swap.
• Reduced motion removes the fly, lift and slide; chips and rows land at once and numbers change in place.

• From 880px the field list sits beside the wells as a column; below it becomes wrapped field pills above the wells.
• Wells sit in a two by two grid from 640px and stack below it, with the well name above its chips on phones.
• The table scrolls inside its card with the header row and first column pinned, and the grand total pinned to the bottom.
• On touch screens chips, fields, row groups and menu items grow to 44px. A tap opens the menu; a drag moves the chip.

• The pivot recomputes in one pass per layout change and memoizes on layout and filters; 1,200 records with nested rows and columns compute in well under a frame.
• Drag hit testing reads layout offsets, not animated rects, so the insert position never oscillates while neighbours move.
• Row and cell animations use transform and opacity only. Very large pivots (hundreds of rows) stay correct but should collapse groups by default.

• Choose it when people need to slice one flat dataset their own way: sales by region and quarter, hours by team and month, tickets by queue and priority.
• Feed records as flat objects keyed by field id. Give dimensions an explicit values order when the natural order is not alphabetical (quarters, months, sizes).
• Start with a meaningful defaultLayout so the first view already answers a question. Use the empty layout only where building from scratch is the point.
• Persist views with onLayoutChange and pass the saved layout back as defaultLayout with a key to restore it.
• The engine is synchronous and client side. Aggregate on the server first for datasets beyond tens of thousands of records.

The full library index for assistants is at /llms.txt.