0004 — `rrxiv.cls` LaTeX template

Status: v0.1 draft. Reference: template/rrxiv.cls. Skeleton paper: template/rrxiv-template.tex. Minimal example: template/examples/minimal/.

Purpose

rrxiv.cls is a thin LaTeX class layered on article that adds:

Required metadata commands — \rrxivid, \rrxivversion, \rrxivprotocolversion, \rrxivlicense, \rrxivtopics. Authors set these once in the preamble.
Six semantic environments — claim, evidence, observation, scope, openquestion, rrxivremark. These render normally in the PDF and emit structured markers in a sidecar file.
Four inline edge declarations — \dependson, \supports, \extendsclaim, \contradicts. These produce no visible output; they only emit sidecar markers.
A sidecar \write channel — every compile produces <basename>.rrxiv.aux containing the markers. The rrxiv parser reads this file to build the CIR.

The class is intentionally minimal. It does not theme the paper, define a custom title page, or impose a citation style. Authors layer their own preferences on top.

Required metadata

Set these once in the preamble:

\rrxivid{my-paper-0001}             % stable paper ID
\rrxivversion{v1}                   % paper revision
\rrxivprotocolversion{0.1.0}        % rrxiv protocol version
\rrxivlicense{CC-BY-4.0}            % SPDX license identifier
\rrxivtopics{topic-1,topic-2}       % comma-separated topic IDs

The class emits these to the sidecar at \begin{document}. They populate the CIR's top-level metadata fields.

Command	Sidecar marker	CIR field
`\rrxivid{X}`	`RRXIV:meta:id:X`	`id`
`\rrxivversion{X}`	`RRXIV:meta:version:X`	`version`
`\rrxivprotocolversion{X}`	`RRXIV:meta:protocol:X`	`rrxiv_version`
`\rrxivlicense{X}`	`RRXIV:meta:license:X`	`license`
`\rrxivtopics{X}`	`RRXIV:meta:topics:X`	`topics` (comma-split)

\rrxivid and \rrxivversion are required. The others have defaults (v1, 0.1.0, CC-BY-4.0, empty topics) but explicit declaration is recommended.

Semantic environments

Environment	Sidecar marker	Purpose	CIR mapping
`claim`	`RRXIV:claim:N`	Single falsifiable assertion	`claims[]`
`evidence`	`RRXIV:evidence:N`	Evidence for the most recent claim	(associated with adjacent claim)
`observation`	`RRXIV:observation:N`	Factual observation, no claim	(an annotation of type `claim_extraction` may promote it)
`scope`	`RRXIV:scope:N`	Conditions under which following claims hold	`claims[].scope`
`openquestion`	`RRXIV:openquestion:N`	Question this paper does not resolve	(annotation target candidate)
`rrxivremark`	`RRXIV:remark:N`	Aside / methodological note	(carried as commentary, not a claim)

N is the LaTeX counter value at \begin{...} time. The parser pairs the Nth sidecar marker of a given kind with the Nth source occurrence of that environment.

Usage

\begin{claim}[A short label for the claim]
\label{claim:my-claim}
State the claim as a single falsifiable assertion in plain language. A reader
should be able to read this paragraph in isolation and know what is being
asserted.
\end{claim}

\begin{evidence}[Evidence for claim:my-claim]
\label{ev:my-evidence}
Sketch or describe the evidence supporting the claim above.
\end{evidence}

The optional argument in [brackets] is a human-readable title shown in the rendered PDF. The \label{...} inside the body is the user-chosen anchor; the parser uses it to construct the canonical claim ID <paper_id>:<label>. If you omit \label, the parser falls back to <paper_id>:claim:N (using the sidecar index).

Recommended labelling convention:

Environment	Label prefix
`claim`	`claim:`
`evidence`	`ev:`
`observation`	`obs:`
`scope`	`scope:`
`openquestion`	`oq:`
`rrxivremark`	`rem:`

These are conventional, not enforced. The parser accepts any unique label.

Inline edges

Declare typed edges between claims with no visible output:

\dependson{my-paper-0001:claim:foo}{other-paper:claim:bar}
\supports{my-paper-0001:claim:foo}{my-paper-0001:claim:baz}
\extendsclaim{my-paper-0001:claim:foo}{some-foundational-paper:claim:base}
\contradicts{my-paper-0001:claim:foo}{another-paper:claim:opposite}

Each emits a RRXIV:edge:<type>:<src>|<dst> marker the parser turns into a graph edge. The IDs follow the <paper_id>:<label> convention. Cross-paper edges are first-class — declare them whenever a claim of yours depends on, supports, extends, or contradicts a claim from another rrxiv paper.

The | separator was introduced in cls v0.2 (per RRP-0002). v0.1 used : to join src and dst, which was ambiguous for IDs that themselves contained colons. The reference parser still accepts the v0.1 format with a DeprecationWarning; recompile any v0.1-era papers with the v0.2 cls to clear the warning.

Compilation

With Tectonic (recommended — single binary, auto-fetches packages):

tectonic --keep-intermediates paper.tex

With a traditional TeX Live install:

pdflatex paper.tex
bibtex paper            # if you have citations
pdflatex paper.tex
pdflatex paper.tex      # second pass for cross-references

You'll get paper.pdf and paper.rrxiv.aux. The sidecar is what the rrxiv parser reads. The PDF is the human rendering target.

Compile dependencies

rrxiv.cls v0.1 loads:

geometry (a4paper, 1in margins; override at \documentclass[]{rrxiv})
inputenc (utf8) and fontenc (T1)
amsmath, amsthm, amssymb, mathtools
graphicx, xcolor, enumitem
natbib (numbered citations)
booktabs
authblk (author affiliations via \affil[]{...})
hyperref (colored links)

These are widely available. A standard TeX Live or Tectonic install has them. If you need to add packages, \usepackage{...} after \documentclass{rrxiv} is fine.

Distributing the class with your paper

When submitting to rrxiv, you upload the source bundle (.tex, .bib, figures, and rrxiv.cls). Don't rely on the server having the class installed. The bundle convention is:

my-paper-0001/
├── my-paper-0001.tex
├── my-paper-0001.bib
├── rrxiv.cls           ← bundled with submission
└── figures/
    └── ...

The submitting tool (rrxiv submit, future) checks that rrxiv.cls is present in the bundle and that its version matches a server-supported range.

v0.1 limitations

Sidecar markers carry only the LaTeX counter index, not the user's label. The parser pairs by ordinal-within-kind; this means rearranging the order of \begin{claim} blocks in source rearranges the implicit ordinals. Use stable \label{...} calls inside each block to anchor identity.
Citations don't emit sidecar markers; the parser scans the source for \cite{...} directly and resolves keys against the .bib file. This is fine for v0.1.
Math mode inside environments is preserved as raw TeX in the CIR's claim statement field. A v0.2 RRP will specify TeX-to-text or TeX-to-MathML for the public-facing renders.
Figures are CIR-internal in v0.1 — they don't participate in the claim graph. A figure can be the target of an annotation (e.g., a dataset_link).

0004 — rrxiv.cls LaTeX template