\section{Summary rows and statistical scope}
\label{sec:summary}
Set aggregate on scalar number, bar, bullet or heat columns and declare one
\key{gradsummary} in the table. Aggregate accepts none (default), count, sum, mean,
min or max. Summary cells display numbers rather than graphics, so a total
outside the drawing range is not clipped.
\begin{demo}{A valid-observation mean and total latency}
\begin{center}
\begin{gradtable}[align=c,header align=c,style=academic,theme=teal,width=\linewidth]
  \gradcolumn{name}{Method}
  \gradcolumn[type=bar,aggregate=mean,mark=both,
    options={max=100}]{score}{Score}
  \gradcolumn[type=number,aggregate=sum,
    options={precision=0,unit={\,ms}}]{time}{Time}
  \gradrow{A,72,20}\gradrow{B,90,30}\gradrow{Missing,NA,0}
  \gradsummary{Summary}
\end{gradtable}
\end{center}
\end{demo}
The score mean is 81 and latency sum is 50. Count counts valid numeric observations
and displays an integer with no unit or computed percent. Missing is skipped;
zero is retained. Empty/all-missing count and sum are 0; mean/min/max are missing.
\subsection{Displayed rows or the full dataset}
\begin{sloppypar}
\key{\gradsummary[scope=visible|data]{label}} defaults to visible, after CSV
filtering, sorting and limit. Data uses all imported rows; both scopes coincide
for manual tables. The label occupies the first text column; use an empty label
for an all-numeric table. Text/compound columns cannot be aggregated and columns
without an aggregate leave empty summary cells.
\end{sloppypar}
\begin{demo}{Display two rows, summarize all imported observations}
\gradbarsloadcsv{results}{docs/data/results.csv}
\begin{center}
\begin{gradtable}[align=c,header align=c,style=compact,data=results,width=\linewidth,
  sort=score,order=descending,limit=2]
  \gradcolumn[source=Model]{name}{Method}
  \gradcolumn[type=number,source=Score,aggregate=mean]{score}{Score}
  \gradcolumn[type=number,width=18mm,source=Latency,aggregate=count]{n}{Latency}
  \gradsummary[scope=data]{All data}
\end{gradtable}
\end{center}
\end{demo}
The full score mean is 61.675; valid latency count is 4. Summaries do not affect
ranking, ranges or source-row counts. Number columns retain their alignment,
bar/bullet summaries align right and heat summaries center. Precision, units
and number format are inherited and measured before width allocation. Long
tables show the summary once with their last data row. Summary color=auto follows
the table theme; an empty color disables fill. Summary font defaults to bold text
and math.

\subsection{Group summaries and independent statistical scopes}
\begin{sloppypar}
\key{group summary=true} adds one numeric row per displayed group using the
existing column aggregates, even without an overall \key{gradsummary}.
Group summary label defaults to Summary in the first text column.
\key{group summary scope=visible|data} defaults to visible; data includes all
imported observations of that group, including filtered/limited-out members.
Undisplayed groups do not acquire headings or summaries. Manual groups have
the same rows in both scopes. Missing, zero and formatting rules match overall
summaries. Group and overall scope settings are independent; color/font reuse
summary color and summary font. Neither summary participates in sorting, ranks
or ranges. The overall summary follows the final group summary on the same page.
\end{sloppypar}
\begin{demo}{Complete example: grouped experiment results}
\gradscale{change}{min=-10,max=10}
\gradbarsloadcsv{experiments}{docs/data/experiments.csv}
\begin{center}
\begin{gradtable}[align=c,header align=c,data=experiments,width=\linewidth,theme=violet,
  group by=batch,group summary=true,group summary label={Group summary},
  sort keys={change/descending,duration/ascending}]
  \gradcolumn[source=Run]{run}{Run}
  \gradcolumn[source=Batch]{batch}{Batch}
  \gradcolumn[type=heat,source=Change,aggregate=mean,scale group=change,
    options={heat palette=diverging,heat center=0}]{change}{Change}
  \gradcolumn[type=number,source=Duration,aggregate=count]{duration}{Duration}
  \gradsummary{All visible}
\end{gradtable}
\end{center}
\end{demo}
The groups have Change means of $-1$ and 3. Duration summaries count valid
observations: two per group and four overall. The overall Change mean is 1.