Interface VegaLiteTransform

Window operations, of which the adapter reads two things.

A running sum is what separates a waterfall from any other ranged bar: both draw a floating bar between y and y2, but a waterfall's two bounds are consecutive running totals, and a running total is a window sum. A ranged bar whose bounds are two measured values (a monthly temperature low and high, say) has no such transform, and announcing its bounds as contributions to a total would invent an accumulation the chart does not draw.

A rank is what separates a bump chart from any other line chart: the y axis carries a place in a table rather than a magnitude, and Vega-Lite has one way to compute one. as is required by the Vega-Lite schema for every window operation, so the ranked column is always named.

The columns a fold transform turns into rows.

That is how a parallel coordinates plot is built: several variables are folded into one key / value pair so a single line mark can draw one polyline per observation across all of them. No Vega-Lite mark says "parallel coordinates"; the fold does.

The field on the primary data a lookup transform joins on.

That field is what identifies a region on a choropleth: the geometry carries a name or a code, the values arrive from a second dataset keyed by it, and the join is the only place a geoshape spec writes the pair down. Vega resolves it as a field accessor, so a dotted name (properties.name) reaches into the feature's own properties.

The field a density transform estimates a distribution over.

The grouping keys of a density or aggregate transform.

The output column names of whichever transform declares them.

fold and density both emit a pair (["key", "value"] and ["value", "density"] by default); calculate and friends emit one.