Academic Writing Skill

Workflow

Step 1 — Read the Paper Structure Map

Always start by reading context/planning/paper_structure.md. This defines what each section must contain and how sections connect. Do not invent section structure; follow the map.

Step 2 — Identify the Section to Write

Determine which section is requested. The expected paper structure is defined in context/planning/paper_structure.md. A typical layout is:

paper/sections/
├── introduction.tex
├── literature.tex
├── model.tex
├── [section_name].tex    # one file per model component / channel
├── ...
├── extensions.tex
├── empirics.tex
└── conclusion.tex

Each section has different source material and different conventions. See the section-specific guidance below.

Step 3 — Write the Section

Write LaTeX that compiles cleanly and reads as top-journal prose. Follow all style rules and LaTeX conventions specified below.

Step 4 — Compression Pass

After writing the section, re-read it with one question: "If I cut this sentence, does the reader lose anything they need?"

Apply these rules before proceeding:

• Delete any sentence that restates a point already made in the same section.
• Replace "It is worth noting that X" with "X."
• If a paragraph has more than two hedge phrases ("may," "potentially," "it is possible that"), remove all but one.
• If a subsection exceeds its target length by more than 20%, cut the weakest paragraphs before proceeding.

Section word-count targets (treat as hard ceilings, not aspirations):

• Introduction: 2,500–3,500 words
• Literature review: 2,500–3,500 words
• Each model component section: 1,500–2,500 words
• Integrating section (if any): 1,500–2,000 words
• Empirics: 1,500–2,500 words
• Conclusion: 800–1,200 words

Step 5 — Verify Consistency

Before finalising any section:

1. Check that every quantitative claim references a specific table, figure, or proposition
2. Check that all \cite{} keys exist (or flag missing entries for the human)
3. Check that notation is consistent with context/model_equations.md
4. Check that cross-references (\ref{}, \eqref{}) point to valid labels
5. Check that the section's argument is consistent with adjacent sections

Section-Specific Guidance

Introduction

Source material: threat_map_final.md (differentiators), research_plan_final.md (contributions), research_context.md (mechanism overview)

Structure:

1. Opening paragraph: the economic problem
2. What this paper does: the model components and their interaction, stated precisely
3. Main results: each contribution as a concrete finding (threshold, comparative static, or characterisation)
4. Related literature: brief positioning (2-3 paragraphs), not a full review
5. Paper organisation: one paragraph mapping sections

Rules:

• State each contribution as a result, not a topic ("We show that..." not "We study...")
• Engage the 2-3 closest threat papers by name with precise differentiators
• Do not oversell; use "characterise", "show", "derive" rather than "prove" or "demonstrate conclusively"
• The integrating contribution (if any) must be framed as the core contribution

Literature Review

Source material: context/literature/review.md (the Literature Guardian's structured prose)

Task: Convert the prose into LaTeX. The analytical structure is already defined in the source file. Do not reorganise or rewrite the argument; format it.

Rules:

• Preserve the section structure from literature_review.md
• Convert all citations to \citet{} (in-text) or \citep{} (parenthetical) as appropriate
• Do not add papers not in the source file
• Do not change threat classifications or differentiator language

Model Sections (model.tex, [section_name].tex files)

Source material: context/model_equations.md, context/research_context.md

Task: Write the prose that surrounds the formal derivations. The equations and propositions come from the Theory Builder; this skill writes the setup, motivation, interpretation, and discussion around them.

Structure per component section:

1. Setup and motivation (1-2 paragraphs): why this component matters, what the foundational model is
2. Model environment: agents, information, timing, written in prose with inline math
3. Equilibrium definition
4. Main result(s): stated as formal propositions in \begin{proposition}...\end{proposition}
5. Comparative statics in the key primitive
6. Discussion: economic interpretation, connection to the next component

Rules:

• All equations in align or equation environments with \label{eq:...}
• All propositions labelled \label{prop:...}
• Notation must match model_equations.md exactly; do not rename variables
• Use LaTeX math commands consistently (never spell out variable names in prose where a math symbol is defined)
• Interpret every mathematical result economically; no unexplained equations

Extensions

Source material: research_plan_final.md (extension descriptions), model_equations.md (if formal results exist)

Rules:

• Frame as natural extensions, not afterthoughts
• Clearly separate from the main model; these are not core contributions

Empirics

Source material: research_plan_final.md, paper/figures/, paper/tables/

Rules:

• Frame as motivating evidence, not causal identification (unless the research design supports causal claims)
• Every number in the text must match a table or figure exactly
• Report statistics with appropriate precision (t-statistics to 2 decimals, coefficients to 3-4 significant figures)
• Use \input{} to include tables generated by the Empirical Agent

Conclusion

Rules:

• Restate contributions concisely (1 sentence each, not a paragraph rehash)
• Policy implications: state concretely
• Limitations: state the scope constraints from research_context.md honestly
• Future work: 2-3 concrete directions, not vague gestures

Style Rules

These rules apply to all sections.

Prose

• Flowing academic prose in paragraphs; no bullet points in the manuscript
• No em-dashes; use commas, semicolons, or parentheses instead
• No bold text within paragraphs (bold is reserved for theorem/proposition labels)
• Active voice for the paper's contributions ("We show that..."); passive voice acceptable for describing others' work or methods
• Present tense for general claims and model properties; past tense for specific empirical findings
• "which" for non-restrictive clauses (with comma); "that" for restrictive clauses (no comma)
• Spell out numbers one through nine; numerals for 10 and above
• First person plural ("we") throughout; never "I" or "the authors"

Citations

• \citet{author2024} when the author is the grammatical subject: "\citet{smith2024} show that..."
• \citep{author2024} for parenthetical references: "...as shown in prior work \citep{smith2024}"
• \citet{author2024a, author2024b} for multiple in-text citations
• Never write author-year citations manually; always use \citet{} or \citep{}

Mathematics

• Display equations: \begin{equation}...\end{equation} for single equations, \begin{align}...\end{align} for multi-line
• Every display equation gets a \label{eq:descriptive-name}
• Inline math for short expressions
• Use \text{} inside math for subscript words: $c_{\text{P}}$ not $c_P$
• Parentheses: \left( and \right) for expressions that vary in height
• Operators: \max, \min, \arg\max, \mathbb{E} (never italic versions)

LaTeX Conventions

• Use \section{}, \subsection{} — never \paragraph{} or manual bold headers
• Tables: \begin{table}[t] with \caption{} above the tabular, \label{tab:...} inside the caption
• Figures: \begin{figure}[t] with \caption{} below, \label{fig:...} inside the caption
• Reference as "Table~\ref{tab:...}" and "Figure~\ref{fig:...}" (capitalised, with non-breaking space)
• Propositions: \begin{proposition}[Short title]\label{prop:...} followed by \begin{proof} if proof is included
• No \textbf{} for emphasis in running text; use italics sparingly or restructure the sentence
• \footnote{} for tangential points; keep footnotes short

Consistency Checks

Before producing any .tex file, verify:

• Every \cite key appears in the BibTeX file (or flag as missing)
• Every \ref and \eqref points to a defined \label
• Notation matches model_equations.md (variable names, subscripts, operator conventions)
• No orphaned claims (quantitative statements without table/figure/proposition reference)
• Section transitions are smooth (the last paragraph of each section should motivate the next)
• The introduction's contribution list matches what the body actually delivers

Output

One or more .tex files in paper/sections/, written to or updated in place.

This skill produces LaTeX source only. It does not compile the document. If the project uses a build system (Quarto, Makefile, justfile), compilation is handled separately.

Anti-patterns

Do not:

• Reorganise the literature review's analytical structure (the Literature Guardian defined it; this skill formats it)
• Invent model results or propositions (those come from the Theory Builder via model_equations.md)
• Use \textbf{} for section-like headers within the body text
• Write bullet-point lists in the manuscript (convert to prose)
• Include raw code, filenames, or pipeline references in the manuscript text
• Use em-dashes anywhere in the manuscript
• Write "as shown above" or "as mentioned earlier" without a specific cross-reference
• Leave \cite{???} placeholders without flagging them for the human