A column whose values have levels: the values it may hold, in an order of their own, which a plot follows wherever it orders the column's values: on an axis, in a legend, in the groups a stat counts and a position stacks, and in a facet's panels.

Put one in a table where a column would go:

%{
  "cut" => Giggly.Factor.new(["Ideal", "Fair"], levels: ["Fair", "Good", "Ideal"]),
  "price" => [326, 327]
}

A factor enumerates as its values, so anything that reads a table (Table.Reader, Explorer's DataFrame.new/1, Kino's data table) reads the column as they are, without its levels.

levels: names each level as the values hold it ([8, 4, 6] for numbers), and a value not among them is missing. Without it, the levels are the distinct values in order: numbers by value, dates and times by time, strings by their bytes, false before true. ordered: true marks the levels as ordered. The table refuses, when a plot reads it, levels that hold nil, that name one value twice (4 and 4.0 are one), or whose type is not the values'.

iex> cut = Giggly.Factor.new(["Ideal", "Fair", "Ideal"], levels: ["Fair", "Good", "Ideal"])
iex> Enum.to_list(cut)
["Ideal", "Fair", "Ideal"]
iex> plot =
...>   %{"cut" => cut}
...>   |> Giggly.Plot.new(%{x: :cut})
...>   |> Giggly.Plot.geom_bar()
iex> Giggly.panel_scales(plot, 1).x.levels
["Fair", "Ideal"]

A level no value holds takes no place on the axis, as Good takes none here.