brycewang-stanford IEEE S&P Writing Style
Use this for drafting and revision passes. The controlling constraint is
S&P's round structure: first-round reviewers cut papers without a
rebuttal (ieeesp-review-process), so the paper's first two pages must
survive a professional skeptic reading alone, unaided, and slightly tired.
The first-round survival test
By the end of page 2, a reviewer must be able to answer:
- Who is the adversary? Capabilities, access, and what winning means.
- What is the claim? One sentence, bounded: "we show an attacker with can on , demonstrated on systems."
- What is the evidence? The map: which section demonstrates, which measures, which proves.
- Why now / why hard? What made this un-done until this paper.
- Was it done responsibly? One early signal — disclosure initiated, IRB obtained — expanded later.
A draft that spends page 1 on the importance of cybersecurity fails this test; Oakland reviewers assume importance and grade specificity.
Threat model as a load-bearing section
The threat model is not scene-setting — every later claim is evaluated against it:
- State capabilities and non-capabilities ("the attacker cannot observe physical access patterns"); reviewers probe boundaries you leave implicit.
- Justify realism with citations to incidents or measurements, not intuition.
- Keep one threat model. Papers that quietly strengthen the adversary in §5 to make the defense interesting get caught in reviewer discussion.
- For defenses: state the adaptive adversary — one who knows the design — and point to where that adversary is evaluated.
Claim calibration table
| Overclaim | Calibrated form |
|---|---|
| "Our defense stops X attacks" | "blocks all X instances in our corpus (n=…); adaptive variants in §6.3" |
| "The first work to…" | "the first demonstration on production systems of…" (and verify) |
| "Devastating real-world impact" | measured impact + affected-population estimate + fix status |
| "Provably secure" | "secure under against (Thm 2)" |
| "Negligible overhead" | "% median overhead on (Fig 7)" |
The pattern: every security adjective is replaced by a boundary plus a pointer. This is also rebuttal insurance — calibrated claims leave less for reviewers to dispute.
SoK voice is a different register
An SoK: paper argues a thesis about the literature: the organizing claim
("defenses in this area implicitly assume X, and the assumption fails")
appears on page 1 like any other claim, the taxonomy is the evidence, and
the systematization must earn its verbs — compare, reconcile,
invalidate — not just survey. Per the CFP, value to the community is the
review criterion; a section explicitly stating what practitioners and
researchers should do differently is the strongest close.
Compression for 13 compsoc pages
Two-column IEEE compsoc punishes sprawl. Cuts that preserve the argument, in order:
- Background that a security-generalist PC member already knows (they know what ASLR is; they need your bypass's delta).
- Repeated attack-flow narration — one figure with a numbered walkthrough replaces a page of prose.
- Per-experiment methodology boilerplate — one Evaluation Setup subsection, referenced thereafter.
- Defensive hedging paragraphs — one Limitations subsection beats hedges scattered through every section.
Revision pass, one command per pass:
pass 1: page-2 test — answers to the 5 questions all present? mark gaps
pass 2: grep -n "first\|novel\|significant\|devastating\|completely" body.tex
→ each hit: boundary + pointer, or delete
pass 3: threat-model diff — list every capability §4–§7 assumes;
confirm each appears in §2
pass 4: read only topic sentences aloud — does the argument survive?
House conventions worth matching
- Present tense for the system and attack ("the attack proceeds in three stages"); past tense for what you measured.
- Third-person self-citation everywhere in the anonymous version.
- Numbered attack stages and named components; reviewers reference "stage 2" and "the monitor" in discussion — make that easy.
- Figures readable in grayscale at print size; compsoc columns are narrow.
Output format
[Page-2 test] adversary ✓/✗ · claim ✓/✗ · evidence map ✓/✗ · why-now ✓/✗ · ethics signal ✓/✗
[Threat model] boundaries explicit? drift across sections? <findings>
[Claim audit] <n> uncalibrated claims with locations>
[Compression] current <n> pp → 13 pp plan: <cut list in order>
[Register] research / SoK — voice matches? <note>