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
- 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.
- The skeptic pass: at every "clearly," "standard," and "it follows,"
either write the missing sentence or verify a graduate student could.
- 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.
- 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>
1---2name: stoc-writing-style3description: Use when drafting or revising a STOC (ACM Symposium on Theory of Computing) paper — the theorem-forward first page, informal/formal statement pairing, the technical-overview section that carries acceptance, notation economy for a cross-area committee, and prose that survives the twelve guaranteed pages.4metadata:5 source: https://skillmd.com/skills/brycewang-stanford/stoc-writing-style6---78# STOC Writing Style910A STOC submission is read by two different people you must satisfy at once: a11program-committee member from a neighboring sub-area deciding whether the result12is a big deal, and a specialist subreviewer deciding whether it is true. STOC's13reading contract (abstract + table of contents + first twelve pages guaranteed;14STOC 2026 CFP, checked 2026-07-08) forces both conversations into the same15window. The house style below is how successful papers manage that.1617## The first page: theorem before context1819Theory flagship papers state the result almost immediately, then earn it.20The standard device is the informal/formal pair:2122```latex23\begin{theorem}[Main result, informal]\label{thm:main-informal}24Every $n$-vertex planar graph admits a spanning structure of weight25$O(\log n)$ times optimal, computable in near-linear time.26\end{theorem}27% Section 2 then states Theorem 2.1, the formal version with the exact28% model, the constant regime, and the failure probability — and the text29% says explicitly: "Theorem 1.1 is the special case of Theorem 2.1 with30% \epsilon = 1/2."31```3233Rules for the pair: the informal statement may simplify but must not lie (no34hidden model changes, no suppressed hypotheses that reverse the interest of the35result); and the formal statement must be findable from it in one hop.3637## The technical overview is the paper3839Modern STOC papers devote two to four of the twelve pages to a section titled40"Technical Overview" or "Our Techniques." This is where acceptance is decided:4142- Narrate the proof as a sequence of *obstacles and moves*: what the natural43 approach is, exactly where it breaks, and what new idea gets past the break.44- Attribute honestly inside the overview — "steps 1–2 follow [XY]; the novelty45 is step 3" reads as confidence, and subreviewers will reconstruct the46 attribution anyway.47- Use one running example or one figure if the construction is geometric;48 cross-area readers hold pictures longer than notation.49- The overview must be faithful to the real proof. An overview describing last50 month's broken strategy while the appendix executes a different one is a51 correctness flag reviewers specifically look for.5253## Notation economy5455| Habit | Why it matters at STOC |56|---|---|57| Introduce notation at first use, never in a front-loaded glossary dump | Cross-area PC members drop off during page-2 symbol walls |58| Prefer words for one-off concepts ("the heavy vertices") over fresh symbols | Every symbol is a tax on the non-specialist reader |59| 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 |60| Reserve display math for statements that get referenced | Inline-display balance is how twelve pages stay readable |61| Standard names for standard things | Renaming expanders or conductance costs goodwill and space |6263## Sentence-level house norms6465- Claims are exact: "we give a $1.99$-approximation" not "we significantly66 improve the approximation ratio." Theory readers convert vague claims to the67 weakest consistent reading.68- "To the best of our knowledge" once, if ever; a related-work section that69 actually knows the literature (`stoc-related-work`) does this job.70- Open problems at the end are a genre convention and a service: two or three71 crisp questions, each with its parameterization pinned down.72- Under double-blind (in force for STOC 2026), self-reference goes third73 person: "extending the technique of [12]" even when [12] is you.7475## Budgeting the twelve pages7677A workable allocation for a single-result paper: one page of introduction and78main statement; one to two pages of context, related work, and open problems;79two to four pages of technical overview; the remainder for preliminaries plus80the proof spine of the main theorem, with routine steps deferred under the81pointer discipline of `stoc-supplementary`. Two smells that the budget failed:8283- **Preliminaries bloat** — three pages of imported definitions before anything84 new happens. Compress by citing standard sources and defining only what the85 overview needs.86- **Statement crowding** — eight theorems in the window, none proved there.87 Committees prefer one theorem they believe over eight they must take on88 faith; demote secondary results to a single summarizing paragraph.8990## Titles and abstracts, theory register9192- Titles state the result when the result is statable: "Undirected93 ST-Connectivity in Log-Space" is the genre's ideal — problem plus resource94 bound, no verbs of enthusiasm. Technique-forward titles ("... via Lazy95 Hierarchies") are the accepted alternative when the technique is the story.96- Self-deprecating precision is house-respected: a title admitting the97 improvement is slight, with the barrier story carried inside, reads as98 confidence (see the exemplars library for a Best Paper that did exactly99 this).100- The abstract carries exact bounds and models, because theory abstracts are101 read as claims: any constant, exponent, or model qualifier in the abstract102 is quotable and will be quoted.103- No citations in the abstract by convention; name prior bounds ("improving104 the best known $O(\log^2 n)$") and cite in the introduction.105106## Revision passes that pay at this venue1071081. **The skim pass:** read only the abstract, ToC, section heads, and theorem109 statements. Is the contribution and its scale unambiguous? That is the110 five-minute PC read.1112. **The skeptic pass:** at every "clearly," "standard," and "it follows,"112 either write the missing sentence or verify a graduate student could.1133. **The breadth pass:** find the three sentences a distributed-computing (or114 crypto, or quantum) PC member needs to care; make sure they exist on page 1.1154. **The consistency pass:** every constant, exponent, and model detail in the116 abstract must match the formal statements exactly — abstract inflation is117 the most commonly caught overclaim.118119## Cycle-volatility warnings120121- The guaranteed-read geometry (and hence all page budgeting above) follows the122 current CFP; the window was 10 pages in earlier cycles and 12 in 2026123 (待核实 each year).124- Anonymization phrasing rules apply only in double-blind cycles; confirm the125 live policy before rewriting self-references.126127## Output format128129```text130[First-page verdict] theorem-forward / buried lede <- fix131[Overview quality] obstacles-and-moves / summary-only <- rewrite132[Notation audit] <symbols to cut, regimes to localize>133[Budget] <pages per role; bloat or crowding found>134[Skim-pass result] <what a 5-minute reader takes away>135```