# Forest Plot — Situational Evidence

Use for comparing point estimates and uncertainty intervals across studies, cohorts, endpoints, models, or subgroups. Do not use for raw distributions, time series, or rankings without uncertainty.

## Required data

Every row needs a label, point estimate, lower bound, upper bound, estimate type or units, and a justified null/reference value. Do not infer intervals, weights, pooled rows, or subgroup membership.

## Required structure

- Treat the figure as an aligned three-column system: label, plot, and formatted estimate.
- Align all content on the same row baseline and keep the columns close enough to trace without heavy guide lines.
- Left-align label and estimate columns by default, including numeric values. Use tabular numerals. Right alignment is allowed when the user asks or a dense statistical convention materially improves comparison.
- Use a log scale and null `1` for ratios such as HR, OR, and RR; use a linear scale and null `0` for signed differences.
- Draw a neutral 1px null line. It is chart structure, not a data series.
- Use one contrast-safe hue from the active visualization profile’s palette for one analysis. Additional colors require a real subgroup, cohort, model, or status meaning.
- Show any interval that exceeds the domain with a continuation marker and retain its full numeric value.

## Marks and intervals

- Use compact square point estimates; use a diamond for a supplied pooled estimate.
- Data-bearing confidence-interval lines may use 1–2px strokes based on final size and density. Keep peer rows equal.
- Marker size may encode supplied study weight only when the size scale is explained.
- Keep exact values in the estimate column rather than labeling interval endpoints in the plot.

## Grouping

- Use whitespace hierarchy before rules or background bands: row gaps are smallest, header-to-child gaps are medium, and subgroup breaks are largest.
- Subgroup headers are label-only rows. Give pooled rows additional breathing room.
- If row tracing remains difficult, first narrow column gutters, then consider subtle neutral row treatment.

## Forbidden

- Do not auto-highlight a “winning” row with bold, blue text, or a surface. Emphasis requires a source- or user-defined reason.
- Do not use red/green unless harmful/beneficial or negative/positive meaning is real and non-color cues are also present.
- Do not silently clip intervals or omit ambiguous uncertainty definitions.
- Do not stretch the three columns across a wide canvas merely because space exists.
- Do not place explanatory prose inside the plot column.

## Final checks

- Estimate, interval, scale, units, precision, and null value are scientifically coherent.
- Rows remain traceable without decorative grid lines.
- The active visualization profile’s palette is used consistently.
- Labels, intervals, null line, and estimate text never collide.
- Directional labels do not claim benefit or harm unless that interpretation is supported.
