focs-writing-style
GitHub指导FOCS论文写作，确保前10页完整呈现核心内容以吸引评审。包括页面预算分配、非正式与形式化定理配对技巧、面向跨领域委员会的行文校准及双盲自引用规范。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill focs-writing-style -g -y
SKILL.md
Frontmatter
{
"name": "focs-writing-style",
"description": "Use when drafting or revising a FOCS (IEEE Symposium on Foundations of Computer Science) paper — making the first ten pages carry the whole case to a broad theory committee, pairing informal and formal theorem statements, keeping single-column 11-point prose readable, and writing double-blind-safe self-references."
}
FOCS Writing Style
FOCS submissions carry no length cap, and the CFP encourages submitting the full version of the paper — but everything outside the abstract, the references, and the first ten pages is read only at the committee's discretion (FOCS 2026 CFP, checked 2026-07-08). Style at this venue is therefore an exercise in front-loading: the opening ten pages must let a committee member from a different corner of TCS decide the paper's fate fairly, even if they never turn to page eleven.
Budgeting the guaranteed pages
A budget that survives contact with real papers:
| Pages | Content | Failure mode it prevents |
|---|---|---|
| 1–1.5 | Problem, headline theorem(s) informally, why they matter | Reviewer cannot state your result after page 2 |
| 1.5–3 | Context: prior bounds, the obstacle, your delta stated precisely | "Novelty unclear" from a non-specialist |
| 3–6 | Technical overview: the proof's actual ideas, with real definitions | "Contribution may be incremental" — ideas were invisible |
| 6–9.5 | Formal statements of all results; proof of one representative lemma in full | Committee cannot check that formality matches the informal claims |
| 9.5–10 | Organization note pointing into the body | Diligent reviewer gets lost after the window |
Every theorem you want credit for must be stated by page ten. A result whose first appearance is page 23 does not exist for most of the committee.
The informal/formal pairing
State each headline result twice: once in the introduction in plain language with all quantifiers honest, once formally with every hypothesis. Keep the two mechanically linked so they cannot drift apart across revisions:
% preamble: restatable theorems, one source of truth
\usepackage{thmtools, thm-restate}
\declaretheorem[name=Theorem]{theorem}
% introduction (inside the ten-page window)
\begin{restatable}[Main]{theorem}{mainthm}\label{thm:main}
For every $\varepsilon>0$ there is a deterministic
$O(m^{1+\varepsilon})$-time algorithm that ...
\end{restatable}
% body or appendix, before the proof
\mainthm*
\begin{proof} ... \end{proof}
The thm-restate discipline is the cheapest insurance against the classic
FOCS embarrassment: an introduction promising a bound the formal theorem
quietly weakens.
Prose calibration for a cross-area committee
- Quantify hedges. "Nearly linear" must resolve to $O(m,\mathrm{polylog},m)$ or $O(m^{1+o(1)})$ within a sentence; the two are different claims and committees notice.
- One new notion per section opening. Single-column 11-point pages hold a lot of math; the reader's constraint is working memory, not ink. Introduce a definition, use it twice, then build on it.
- Confession phrases are debts. Every "it is standard that", "one can verify", and "similarly" in the guaranteed pages will be audited by someone; each must either point at a numbered reference or be provable by a reader in under a minute.
- Name your techniques. A committee remembers "the lazy-recoloring argument" and forgets "our second technical contribution". Naming is not vanity; it is compression for the PC discussion.
- Write the obstacle paragraph. The single highest-leverage paragraph in a FOCS submission explains why the previous best approach could not give your result. It converts "difficult" from an adjective into an argument.
Double-blind prose
FOCS has used double-blind review for several cycles (2023–2026 CFPs all require it), with the stated purpose of enabling an unbiased first judgment — not making authorship undiscoverable. The prose consequences:
- Self-citations in the third person: "Chen et al. showed", never "we previously showed" — even when the antecedent is your own arXiv posting.
- No acknowledgements, grant numbers, or lab-specific phrasing ("our group's earlier framework") at submission time.
- Citing your own public full version is normal in this community; what is banned is prose that asserts the identity, not literature that permits a determined guess.
Abstract anatomy
The abstract is guaranteed reading and often the only text a non-assigned PC member sees before the discussion. A FOCS abstract that does its job contains, in order: the problem in one sentence; the main bound or theorem with honest quantifiers; the one-line comparison to the previous best; the technique named in a phrase; and, when the paper has them, the downstream consequences. What it omits: motivation boilerplate ("X is a fundamental problem"), citations, and any claim the formal theorems do not support. Write it last, from the restatable statements, so the abstract inherits their honesty.
Notation economy
Single-column pages at 11 point invite dense macro-heavy notation; committees across subareas punish it. Working rules:
- Prefer words to fresh symbols for anything used fewer than three times; a fourth-order tensor of indexed families needs a name, an ad-hoc set does not.
- Keep one notational convention per object class through the whole paper —
a graph that is $G$ on page 2 must not become $\mathcal{G}$ on page 30 of
the full version (statement-freeze tooling in
focs-reproducibilitycatches drift in theorems, not prose; only discipline catches this). - Standard objects get standard symbols; novelty budget is spent on results, not on renaming expansion constants.
Revision pass order
- Ten-page audit: print pages 1–10 alone; have a colleague reconstruct all claimed results from them.
- Quantifier audit: every informal statement checked against its restatable twin.
- Confession audit: grep the LaTeX for hedge phrases and discharge each.
- Blind audit: third-person self-references, metadata, acknowledgements.
Do the passes in that order — narrative surgery invalidates later audits, so anonymity always goes last.
Version History
- 9f86f09 Current 2026-07-19 15:45
