\part{Graphics and comparisons}

\section{Signed bars, targets, intervals and uncertainty}
\begin{demo}{A signed point estimate and asymmetric uncertainty}
\gradbar[min=-50,max=100,width=65mm,
  target=40,band={20,60},error minus=8,error plus=12]{35}

\gradbar[min=-50,max=100,width=65mm,shape=lollipop]{-25}

\gradrange[min=-50,max=100,width=65mm,range point=10]{-20}{35}
\end{demo}
Error magnitudes are nonnegative distances from the observation, not absolute
endpoints. Explain whether they represent standard deviation, standard error or
confidence intervals in the table caption. Targets and bands are references, not
automatically estimated uncertainty. A band must have ordered endpoints within
the scale; invalid reference bands and targets cause errors.

Floating intervals require lower$\le$upper; equal endpoints draw a cap, not an
inflated bar. Both endpoints must be present or both missing. A range point must
lie inside the original endpoints. Intervals reject error whiskers and quality
thresholds. Default labels show the supplied endpoints.
Lollipops show a point for zero, but no point for missing input.

Negative bars are solid by default. To give the negative side its own gradient,
set its two endpoints explicitly:
\begin{demo}{Independent negative gradient}
\gradbar[min=-100,max=100,width=50mm,
  left color=teal,right color=teal!25,
  negative fill=gradient,
  negative left color=magenta!25,
  negative right color=magenta]{-60}
\end{demo}
The negative left endpoint is the outer side of the bar and the right endpoint is
the side next to zero. `negative color` remains the fallback solid color; threshold
and print modes continue to override the gradient.