# Bar Chart — Situational Evidence

Use for comparing magnitude across discrete categories when values share a meaningful baseline. Use a line chart for continuous progression and a scatter plot for relationships between quantitative variables.

## Required

- Start the quantitative axis at zero. If the comparison cannot tolerate a zero baseline, use a dot plot or another form.
- Use one contrast-safe stop from the active visualization profile’s palette per series or documented category meaning.
- Keep the categorical order meaningful: source order for sequences, sorted order for rank, or grouped order for a real taxonomy.
- Show category labels and units. Use direct values only when the chart must support exact reading.
- Keep peer bars the same width. Preserve visible gutter between grouped bars and a larger gap between groups.
- Round only the value end. Keep the baseline end and internal stacked joins square.

## Allowed choices

- Choose vertical bars for short labels and modest category counts; choose horizontal bars for long labels, rank, or many categories.
- Use grouped bars for two to four peer series. Beyond that, prefer small multiples, a dot plot, or a table.
- Use stacked bars only when the total and part-to-whole composition matter. Use a 100% stack for proportional composition.
- Use restrained value-end radii from the existing radius tokens; tiny marks may remain square.
- Add an error bar when uncertainty is supplied and its definition is available.

## Color and legends

- A single series uses one color across all bars unless another real variable is encoded.
- Multiple series sample distinct stops from the active visualization profile’s palette and require a legend or direct labels.
- Use status color only in product charts when a bar denotes a real product status. Scientific bars retain their scientific encoding.
- Pattern may reinforce selection, uncertainty, or accessibility and must appear in the legend.

## Forbidden

- Do not color every bar differently for variety.
- Do not use gradients, shadows, glows, bevels, or strokes around every bar.
- Do not use stacked bars for precise comparison of internal segments without a shared baseline.
- Do not place labels inside bars that are too narrow or fail contrast.
- Do not add a chart when a nearby table already answers the same question and the visual shape adds no insight.

## Final checks

- The baseline is zero and visually clear without becoming heavier than the data.
- Category and group gaps communicate the grouping correctly.
- All series colors come from the active visualization profile’s palette.
- Error bars state whether they represent SD, SEM, CI, or another statistic.
- Labels and values remain legible at final size.
