Stoc writing style

STOC Writing Style

A STOC submission is read by two different people you must satisfy at once: a program-committee member from a neighboring sub-area deciding whether the result is a big deal, and a specialist subreviewer deciding whether it is true. STOC's reading contract (abstract + table of contents + first twelve pages guaranteed; STOC 2026 CFP, checked 2026-07-08) forces both conversations into the same window. The house style below is how successful papers manage that.

The first page: theorem before context

Theory flagship papers state the result almost immediately, then earn it. The standard device is the informal/formal pair:

\begin{theorem}[Main result, informal]\label{thm:main-informal}
Every $n$-vertex planar graph admits a spanning structure of weight
$O(\log n)$ times optimal, computable in near-linear time.
\end{theorem}
% Section 2 then states Theorem 2.1, the formal version with the exact
% model, the constant regime, and the failure probability — and the text
% says explicitly: "Theorem 1.1 is the special case of Theorem 2.1 with
% \epsilon = 1/2."

Rules for the pair: the informal statement may simplify but must not lie (no hidden model changes, no suppressed hypotheses that reverse the interest of the result); and the formal statement must be findable from it in one hop.

The technical overview is the paper

Modern STOC papers devote two to four of the twelve pages to a section titled "Technical Overview" or "Our Techniques." This is where acceptance is decided:

• Narrate the proof as a sequence of obstacles and moves: what the natural approach is, exactly where it breaks, and what new idea gets past the break.
• Attribute honestly inside the overview — "steps 1–2 follow [XY]; the novelty is step 3" reads as confidence, and subreviewers will reconstruct the attribution anyway.
• Use one running example or one figure if the construction is geometric; cross-area readers hold pictures longer than notation.
• The overview must be faithful to the real proof. An overview describing last month's broken strategy while the appendix executes a different one is a correctness flag reviewers specifically look for.

Notation economy

Habit	Why it matters at STOC
Introduce notation at first use, never in a front-loaded glossary dump	Cross-area PC members drop off during page-2 symbol walls
Prefer words for one-off concepts ("the heavy vertices") over fresh symbols	Every symbol is a tax on the non-specialist reader
State parameter regimes next to results, not in a global convention	"Throughout, $\varepsilon < 1/\log n$" declared on page 3 and used on page 40 will be missed
Reserve display math for statements that get referenced	Inline-display balance is how twelve pages stay readable
Standard names for standard things	Renaming expanders or conductance costs goodwill and space

Sentence-level house norms

• Claims are exact: "we give a $1.99$-approximation" not "we significantly improve the approximation ratio." Theory readers convert vague claims to the weakest consistent reading.
• "To the best of our knowledge" once, if ever; a related-work section that actually knows the literature (stoc-related-work) does this job.
• Open problems at the end are a genre convention and a service: two or three crisp questions, each with its parameterization pinned down.
• Under double-blind (in force for STOC 2026), self-reference goes third person: "extending the technique of [12]" even when [12] is you.

Budgeting the twelve pages

A workable allocation for a single-result paper: one page of introduction and main statement; one to two pages of context, related work, and open problems; two to four pages of technical overview; the remainder for preliminaries plus the proof spine of the main theorem, with routine steps deferred under the pointer discipline of stoc-supplementary. Two smells that the budget failed:

• Preliminaries bloat — three pages of imported definitions before anything new happens. Compress by citing standard sources and defining only what the overview needs.
• Statement crowding — eight theorems in the window, none proved there. Committees prefer one theorem they believe over eight they must take on faith; demote secondary results to a single summarizing paragraph.

Titles and abstracts, theory register

• Titles state the result when the result is statable: "Undirected ST-Connectivity in Log-Space" is the genre's ideal — problem plus resource bound, no verbs of enthusiasm. Technique-forward titles ("... via Lazy Hierarchies") are the accepted alternative when the technique is the story.
• Self-deprecating precision is house-respected: a title admitting the improvement is slight, with the barrier story carried inside, reads as confidence (see the exemplars library for a Best Paper that did exactly this).
• The abstract carries exact bounds and models, because theory abstracts are read as claims: any constant, exponent, or model qualifier in the abstract is quotable and will be quoted.
• No citations in the abstract by convention; name prior bounds ("improving the best known $O(\log^2 n)$") and cite in the introduction.

Revision passes that pay at this venue

1. The skim pass: read only the abstract, ToC, section heads, and theorem statements. Is the contribution and its scale unambiguous? That is the five-minute PC read.
2. The skeptic pass: at every "clearly," "standard," and "it follows," either write the missing sentence or verify a graduate student could.
3. The breadth pass: find the three sentences a distributed-computing (or crypto, or quantum) PC member needs to care; make sure they exist on page 1.
4. The consistency pass: every constant, exponent, and model detail in the abstract must match the formal statements exactly — abstract inflation is the most commonly caught overclaim.

Cycle-volatility warnings

• The guaranteed-read geometry (and hence all page budgeting above) follows the current CFP; the window was 10 pages in earlier cycles and 12 in 2026 (待核实 each year).
• Anonymization phrasing rules apply only in double-blind cycles; confirm the live policy before rewriting self-references.

Output format

[First-page verdict] theorem-forward / buried lede <- fix
[Overview quality] obstacles-and-moves / summary-only <- rewrite
[Notation audit] <symbols to cut, regimes to localize>
[Budget] <pages per role; bloat or crowding found>
[Skim-pass result] <what a 5-minute reader takes away>