% A field guide only; the benchmark is deliberately a separate document.
\section{The Benchmark Suite}\label{ch:rmp}

The manual teaches the language.  The separate \emph{tenkz RMP benchmark}
stress-tests it against 130 agreed targets: 92 published targets in paper
order and 38 author-workbench targets in source order.  It is built from
\tnfile{docs/tenkz/rmp-benchmark.tex}; this manual does not duplicate its
target catalogue.

Each target lives in one preamble-free source file under
\path{tests/tenkz/rmp/}.  That exact file is wrapped for standalone
regression, input by the book, rendered by the web/SVG build, and printed
verbatim on the target's spread.  No source body is copied into a driver,
manifest, or manual.

Every target header states its formula, ink, boundary, source anchors, and
capabilities.  The verso page prints that contract and complete line-numbered
source.  The recto page gives the vector rendering, audited boundary
signature, dimensions, digests, and print/web verdict.  Digest changes make a
prior verdict stale automatically.

One script drives every check:

\begin{Verbatim}[fontsize=\footnotesize,frame=leftline]
python3 scripts/tenkz_rmp.py check --id <target>
python3 scripts/tenkz_rmp.py check --section <section>
python3 scripts/tenkz_rmp.py check --all
python3 scripts/tenkz_rmp.py book --all
python3 scripts/tenkz_rmp.py render --all
python3 scripts/tenkz_rmp.py compare --all --source-root /absolute/path/to/author-source-tree
\end{Verbatim}

The benchmark style controls typography only.  It does not extend tenkz.
The separate author-source tree is not part of this repository.  Reference
comparisons are compiled temporarily from its mapped source blocks; prebuilt
PDFs are not design evidence.  Targets without author drawing source do not
pretend that a reference implementation exists.  They carry the status
\texttt{paper-context-}\allowbreak\texttt{only}.