Stat Paper Write

Stat Paper Write: Section-by-Section LaTeX Generation

Draft a statistics, applied statistics, or ML theory LaTeX paper based on: $ARGUMENTS

Constants

• CLAUDEREVIEWERMODEL = claude-opus-4-6 — Claude model invoked via the Agent tool for internal subagent review.
• CODEXREVIEWERMODEL = gpt-5.5 — External LLM invoked via Codex MCP for senior-statistician external review at model_reasoning_effort: xhigh.
• REVIEW_MODE = both — Options: claude, codex, both. Default both for journal submissions. Switch to claude for quick drafting iteration.
• SUPPLEMENT_MODE — controls how the skill writes supplementary material and how the main paper refers to it.
• separate_self_contained: the main paper and supplement are separate submission artifacts. Do not use cross-file LaTeX references. Cite the supplement textually or bibliographically from the main paper. Make the supplement readable on its own.
• linked_appendix: the supplement is expected to be explicitly referenced from the main paper, or the venue normally treats appendix/supplement linkage as part of the workflow. Cross-references from the main paper to the supplement are allowed when the venue supports them.
• Default SUPPLEMENT_MODE mapping used by this skill family:
• separate_self_contained: AOS, AOAS, BERNOULLI, BIOMETRIKA, EJS, JASA, JASA_ACS, JRSSB, STATSINICA, MSL, JCGS, JABES, STATMED
• linked_appendix: BIOSTATISTICS, COLT, ALT
• Override rule: these are defaults, not absolute claims. If the user provides current journal instructions or a venue template that clearly supports a linked appendix, override the default.
• TARGET_VENUE = AOS — Default venue. Supported:
• Theory/methodology: AOS, JASA, JRSSB, BIOMETRIKA, BERNOULLI, EJS, STATSINICA, MSL
• Application: AOAS, JASA_ACS (alias JASA_APP), BIOSTATISTICS, STATMED, JCGS, JABES
• ML theory conferences: COLT, ALT
• PAPER_TYPE = auto — theory, methodology, application, or auto.
• ANONYMOUS = false — Most stat journals are NOT anonymous. Set true for COLT/ALT conference submissions.
• DBLP_BIBTEX = true — Fetch real BibTeX from DBLP/CrossRef/MathSciNet.

Inputs

1. PAPER_PLAN.md — outline with Theorems-Evidence Matrix, section plan, assumption map (from /stat-paper-plan)
2. NARRATIVE_REPORT.md — the research narrative
3. Generated figures — PDF/PNG files in figures/
4. LaTeX includes — figures/latex_includes.tex (from /paper-figure)
5. Bibliography — existing .bib file, or will create one

If no PAPER_PLAN.md exists, ask the user to run /stat-paper-plan first or provide a brief outline.

Writing References

Read these shared references when they improve writing quality:

• Read ../stat-shared-references/stat-positioning-and-claims.md before drafting the abstract, introduction, contribution list, theorem statements, related work, or discussion. This is the primary defense against the two most common Big Four rejection drivers: weak positioning and overclaim. The reference describes the positioning audit, the technical claim strength audit, and the CLAIM_SUPPORT_MAP.md artifact.
• Read ../stat-shared-references/stat-writing-principles.md before drafting the Abstract, Introduction, or when prose needs statistics-specific voice.
• Read ../stat-shared-references/stat-style-discipline.md during every drafting and clarity pass. This is the primary defense against AI-shaped prose. It covers punctuation rules (no em-dash, no colon, reduce semicolons), AI-template patterns to remove, paragraph and bullet discipline, and COPSS-style scholar writing patterns.
• Read ../stat-shared-references/stat-figure-design.md before generating or polishing any figure. It covers the no-title rule, caption discipline, legend placement, sizing, and overlap prevention.
• Read ../stat-shared-references/stat-venue-checklists.md during setup and final checks.
• Read ../stat-shared-references/stat-theory-writing.md when writing assumptions, theorems, proof sketches, or rate comparison tables (especially for theory and methodology papers).
• Read ../stat-shared-references/stat-application-writing.md when writing an application paper (PAPER_TYPE = application), especially the Data and Background section and the Application / Real Data Analysis section.
• Read ../shared-references/citation-discipline.md when the DBLP/CrossRef workflow is insufficient.

Templates

Venue-Specific Setup

AOS / BERNOULLI / EJS / AOAS (IMS journals):

\documentclass[aos]{imsart}  % or [bj], [ejs], [aoas]
\usepackage{natbib}
\bibliographystyle{imsart-nameyear}
% Author block:
\begin{aug}
\author{\fnms{First} \snm{Last}\ead[label=e1]{email@univ.edu}}
\address{Department, University, City, Country}
\runauthor{Last et al.}
\runtitle{Short Title}
\end{aug}

For AOAS (Annals of Applied Statistics), use \documentclass[aoas]{imsart} and additionally include subject area classification.

JASA Theory and Methods and Applications and Case Studies (both tracks):

Use the official JASA / Taylor & Francis LaTeX template linked from the JASA Instructions for Authors page; check https://www.tandfonline.com/action/authorSubmission?show=instructions&journalCode=uasa20 at submission time. The bibliography style file (.bst) should be the one bundled with the official template.

JASA submission preamble:

\documentclass[12pt]{article}
\usepackage[margin=1in]{geometry}         % 1-inch margins
\usepackage{amsmath,amssymb,amsthm}
\usepackage{graphicx,booktabs,natbib,caption,subcaption,setspace}
\usepackage[hidelinks]{hyperref}
\usepackage{bm,xcolor,microtype,threeparttable,multirow,mathtools}

\theoremstyle{plain}
\newtheorem{theorem}{Theorem}
\newtheorem{proposition}{Proposition}
\newtheorem{lemma}{Lemma}
\newtheorem{corollary}{Corollary}
\theoremstyle{definition}
\newtheorem{definition}{Definition}
\theoremstyle{remark}
\newtheorem{remark}{Remark}

\doublespacing            % JASA: fully double-spaced (required)
\bibliographystyle{agsm}   % ASA-journal helper templates use agsm.bst;
                            % override with the JASA-bundled .bst when available
% Author-year citations: \citet{}, \citep{}
% Submission category: "Theory and Methods" (T&M) or "Applications and Case Studies" (ACS)

Required formatting: 12-point font and fully double-spaced manuscript (\doublespacing from the setspace package). Drafts using \onehalfspacing or LaTeX default single spacing are non-compliant. The 35-page limit assumes fully double-spaced pages.

Length: 35 double-spaced pages for the main manuscript (roughly 26-27 lines per page) as the operative working limit. Move proofs, technical lemmas, extended simulations, and overflow tables and figures to the supplement. Verify the counting boundary on the live Instructions for Authors page before submission.

Abstract: JASA abstracts target 200-250 words. Long drafts of 300-400 words should be cut to this range before submission. Drop the second motivating paragraph (the introduction carries that material), and merge result-and-implication sentences.

Reproducibility (JASA, both tracks):

• Complete the Author Contributions Checklist (ACC) form (https://jasa-acs.github.io/repro-guide/pages/acc.html) and upload it as supplementary material at initial submission.
• A reproducibility-package skeleton (https://github.com/jasa-acs/repro-template) is suggested but not required. Using it makes the revision-stage reproducibility review easier.

Public helper templates if useful (treat as convenience, not source of record):

• ASA-journal Quarto template: https://github.com/quarto-journals/jasa

See ../stat-shared-references/stat-venue-checklists.md for the full JASA block including AI disclosure, peer-review anonymity, alt text, and cover-letter expectations.

Biostatistics (Oxford):

% Use Biostatistics template (oup style)
\documentclass{article}
\usepackage{natbib}
\bibliographystyle{biostatistics}  % or per journal guidelines
% Biomedical / health applications with statistical innovation

Statistics in Medicine (Wiley):

% Use Wiley NJD class or basic article
\documentclass{article}
\usepackage{natbib}
\bibliographystyle{wileyNJD-AMA}  % or per journal guidelines

JCGS (Journal of Computational and Graphical Statistics):

% Use JCGS / Taylor & Francis template
\documentclass[12pt]{article}
\usepackage{natbib}
\bibliographystyle{asa}
% Strong emphasis on reproducibility — code repository must be linked

JRSS-B:

% Use JRSS-B template from journal website
\documentclass[12pt]{article}
\usepackage{natbib}
% Custom JRSS bibliography style

BIOMETRIKA:

% Use Biometrika template
\documentclass{article}
\usepackage{natbib}
\bibliographystyle{biometrika}

STATSINICA:

% Use Statistica Sinica template
\documentclass[12pt]{article}
\usepackage{natbib}

COLT:

\documentclass[anon]{colt2026}  % anonymous submission
\usepackage{natbib}
% Anonymous author block

ALT:

% Use ALT/PMLR template
\documentclass{article}
\usepackage{natbib}

Project Structure (Theory/Methodology Paper)

paper/
├── main.tex                    # master file
├── imsart.cls                  # or venue-specific class file
├── imsart-nameyear.bst         # or venue-specific bst
├── math_commands.tex           # shared math macros
├── references.bib              # bibliography (only cited entries)
├── sections/
│   ├── 0_abstract.tex
│   ├── 1_introduction.tex
│   ├── 2_setup.tex             # problem setup, model, assumptions
│   ├── 3_main_results.tex      # theorems, rate comparisons
│   ├── 4_method.tex            # estimator/algorithm (if separate)
│   ├── 5_simulations.tex
│   ├── 6_real_data.tex         # optional
│   └── 7_discussion.tex
├── supplement/
│   ├── supplement_main.tex     # master file for supplement
│   ├── A_proofs_main.tex       # proofs of main theorems
│   ├── B_proofs_lemmas.tex     # proofs of technical lemmas
│   ├── C_additional_sims.tex   # additional simulations
│   └── D_additional_data.tex   # additional data analysis
└── figures/

Project Structure (Application Paper)

For application papers, the structure shifts to emphasize data and the application. Section files are renamed and reweighted:

paper/
├── main.tex                    # master file
├── imsart.cls                  # or venue-specific class file (use [aoas] for AOAS)
├── math_commands.tex
├── references.bib
├── sections/
│   ├── 0_abstract.tex
│   ├── 1_introduction.tex
│   ├── 2_data.tex              # Data and Scientific Background (2-4 pages)
│   ├── 3_method.tex            # Methodology, scoped to the problem
│   ├── 4_theory.tex            # Light theory (1-2 theorems)
│   ├── 5_simulations.tex       # DGPs informed by the real data
│   ├── 6_application.tex       # Application / Real Data Analysis (4-6 pages — LARGEST)
│   └── 7_discussion.tex        # with practical recommendations
├── supplement/
│   ├── supplement_main.tex
│   ├── A_data_details.tex      # extended data description, preprocessing
│   ├── B_proofs.tex             # proofs (moved to supplement for app papers)
│   ├── C_additional_sims.tex
│   ├── D_additional_analyses.tex  # sensitivity, subgroup, robustness
│   └── E_reproducibility.tex   # software, computation, replication notes
├── figures/                     # rich EDA + analysis figures
└── data/                        # data availability docs or links

For application papers, the figures directory and §2/§6 are where the paper's quality is judged. Plan accordingly.

Section files are FLEXIBLE: Match the paper plan structure. A theory paper may have 6 sections; a methodology paper may have 8; an application paper typically has 7 with §6 dominating.

Main and Supplement Separation

The handling of the supplement depends on SUPPLEMENT_MODE (see Constants above).

For SUPPLEMENT_MODE = separate_self_contained (default for JASA, AoS, AOAS, JRSS-B, Biometrika, EJS, Bernoulli, Statistica Sinica, JASA ACS, JCGS, JABES, Statistics in Medicine, MSL):

The main paper and supplement are independent PDFs uploaded separately. The supplement is a self-contained document.

For SUPPLEMENT_MODE = linked_appendix (default for Biostatistics, COLT, ALT):

The supplement is part of the same review workflow, and cross-references between the main paper and the supplement are expected. Use \ref across files when the venue's submission system supports it.

The rest of this section describes separate_self_contained, which is the more demanding case. The linked_appendix case relaxes these rules.

The supplement must be a self-contained document. It cannot rely on \ref, \eqref, \cite to labels that live only in the main paper, because the main paper is not compiled when the supplement is compiled.

What this means in practice:

1. The supplement must restate every theorem, lemma, assumption, and equation it proves. Do not write Proof of Theorem 1 and assume the reader knows what Theorem 1 says. Restate it.

1. The supplement must redefine notation and assumptions it uses. Either redefine inline or include a brief notation section at the start of the supplement.

1. Avoid cross-references between main and supplement files. Instead, use textual references that remain readable even when the cross-reference does not resolve:

• In the main paper: "the full proof is given in Section A of the supplement" rather than \ref{sec:proof} to a label in the supplement
• In the supplement: "Theorem 1 (restated from the main paper)" rather than \ref{thm:main} to a label in the main paper

1. Compile the supplement as a separate document. Use \documentclass again at the start of the supplement file, with the same class as the main paper. The supplement should produce its own title page (e.g., "Supplementary Material for [paper title]").

1. Bibliography: each file has its own bibliography. They can share the same .bib file, but each must \bibliography{references} independently.

1. Page numbering: the supplement should restart page numbering at 1, often using S1, S2, ... (capital S prefix).

1. Equation, theorem, figure, table numbering: prefix with S in the supplement. For example, Theorem S.1, Equation (S.3), Figure S.2.

Recommended supplement preamble pattern:

\documentclass[aos]{imsart}  % match main paper class
\usepackage{...}             % same packages
\input{math_commands.tex}    % shared math notation

% Renumber with S prefix
\renewcommand{\thesection}{S.\arabic{section}}
\renewcommand{\thetheorem}{S.\arabic{theorem}}
\renewcommand{\theequation}{S.\arabic{equation}}
\renewcommand{\thefigure}{S.\arabic{figure}}
\renewcommand{\thetable}{S.\arabic{table}}
\setcounter{page}{1}

\begin{document}
\title{Supplementary Material for `[Paper Title]'}
\author{[same authors as main paper]}
\maketitle

\section*{Overview}
This supplement contains: (i) full proofs of Theorems 1--3 from the main
paper, restated for convenience; (ii) additional simulation results;
(iii) [...].

\section{Proof of Theorem 1}
\label{sec:supp:thm1}
\textit{Theorem 1 (from main paper).} [restate the theorem statement here]

\textit{Proof.} [...]

In the main paper, refer to supplement results textually:

% In the main paper:
Theorem 1 establishes the rate of convergence; the full proof,
together with additional technical lemmas, is given in Section S.1
of the Supplement.

In the supplement, do not write subsection headings like \subsection{Proof of Theorem~\ref{thm:saturation}} if thm:saturation is a label defined in the main paper. When the supplement compiles standalone, that \ref becomes ?? and the heading reads "Proof of Theorem ??". Instead, use a textual re

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: gyf9712
• Source: gyf9712/stat-writing-skills
• License: MIT

Install and usage instructions live in the source repository linked above.