INFOCOM Writing Style
Use this when revising the main paper. INFOCOM papers are read by a broad networking TPC that
spans theory and systems, inside a nine-page IEEEtran budget and (traditionally) with no
rebuttal. Two constraints follow: the paper must be legible to both an analyst and a systems
reader, and it must pre-empt the reviewer's objection in the text, because you will not get to
answer it later.
Revision rules
- State the networking problem and the system model first. By the end of the introduction a
reader should know the setting (topology, traffic, constraints), the gap, the contribution
(analysis and/or design), and what improves. Do not open with a technology-trend paragraph.
- Make the model do work. If you formulate an optimization or a Markov/queueing model, its
assumptions must be stated plainly and its results must yield insight or an algorithm — not a
formalism that the evaluation ignores. Reviewers reward a model whose theorem changes what you
build or bound.
- Pair every claim with proportional evidence — a proof for an analytical claim, a
simulation/testbed number with the setup named, or a measurement with its methodology — not
adjectives like "significantly improves."
- Write defensively. Name the assumption a theory reviewer will question and justify it; name
the baseline a systems reviewer will call unfair and tune it; state the regime where your result
does not hold. Because there is no rebuttal, an unaddressed objection becomes a reject.
- Respect the page budget as a design constraint. Nine pages of text (appendices included) is
tight; a proof sketch in the body with the full proof compressed, and a focused evaluation, beat
an over-scoped paper that only fits by cutting the model.
- Maintain double-blind wording in self-citations (third person), tool names, testbed/cluster
identifiers, acknowledgements, and funding.
Networking-paper skeleton
| Section |
Job it must do |
Common failure |
| Intro |
Problem, setting, gap, contribution, what improves — by the first column or two |
Leads with a trend, not a networking problem |
| System model |
Topology, traffic, constraints, assumptions, notation, stated plainly |
Assumptions hidden or introduced only in the evaluation |
| Analysis / design |
Theorems + proof sketches, or the protocol/algorithm, reproducibly |
A model no result uses; a design too thin to re-implement |
| Evaluation |
Each claim answered with proportional evidence, honest baselines |
Toy topology; untuned baseline; metric that proxies the claim |
| Discussion / limits |
The regime where the result holds; deployment caveats |
No limits stated (invites the objection you cannot rebut) |
| Related work |
Delta-first positioning against the networking literature |
Citation catalog with no contrast |
Sentence-level rewrites
| Draft pattern |
INFOCOM-safe rewrite |
| "Our scheme significantly improves throughput." |
"improves median throughput by X% (95% CI ...) over in <topology/scale>" |
| "We prove our algorithm is optimal." |
"We prove a (1-1/e) approximation under ; §V bounds the gap empirically" |
| "We evaluate on a large simulation." |
"We simulate in <ns-3/custom>, , ; setup in §V-A" |
| "The model captures the system." |
"Under assumptions A1-A3 (stated and justified in §III), the model yields Theorem 1" |
| "State-of-the-art performance." |
Claim scoped to the topology, traffic, and regime actually tested |
Defensive-writing discipline (no rebuttal)
[Assumption] which modeling assumption will a theory reviewer challenge? -> justify it in §III
[Baseline] which comparison will a systems reviewer call unfair? -> tune it, document the budget
[Scale] does the evaluation reach a realistic size? -> if not, scope the claim and say so
[Regime] where does the result break? -> state it before a reviewer finds it
-> each objection answered in the PDF, because there is no turn to answer it in a reply
Vignette: compressing an over-scoped analytical paper
A draft with two models, five theorems, and a sprawling evaluation: keep the one model the design
uses, the two theorems that carry the insight (full proofs compressed or sketched), and the two
evaluation figures that test the headline claim; move notation-heavy derivations into a tight
in-budget appendix (remember it counts toward the nine pages); cut the second model to a remark.
The test of a good cut: a reviewer should be able to state your assumptions, your main result, and
why your baseline is fair — from the body alone.
Output format
[Writing diagnosis] clear / under-modeled / over-claimed / evidence-mismatched / over-scoped
[First-page fix] <new framing leading with the networking problem + system model>
[Claim audit] <claim -> proof or simulation/measurement -> proportional? yes/no>
[Defensive gaps] <assumption/baseline/scale/regime objection -> where to pre-empt it>
[Anonymity edits] <self-citations / tool / testbed names / acks to rewrite>
1---2name: infocom-writing-style3description: Use when revising an IEEE INFOCOM paper for a networking problem and system model stated up front, theorems or protocol design that pay off, evaluation proportional to the claim, defensive writing that pre-empts objections (because there is no rebuttal), double-blind wording, and disciplined use of the tight IEEEtran two-column page budget.4---56# INFOCOM Writing Style78Use this when revising the main paper. INFOCOM papers are read by a **broad networking TPC** that9spans theory and systems, inside a **nine-page IEEEtran budget** and (traditionally) with **no10rebuttal**. Two constraints follow: the paper must be legible to both an analyst and a systems11reader, and it must **pre-empt the reviewer's objection in the text**, because you will not get to12answer it later.1314## Revision rules1516- **State the networking problem and the system model first.** By the end of the introduction a17 reader should know the setting (topology, traffic, constraints), the gap, the contribution18 (analysis and/or design), and what improves. Do not open with a technology-trend paragraph.19- **Make the model do work.** If you formulate an optimization or a Markov/queueing model, its20 assumptions must be stated plainly and its results must yield insight or an algorithm — not a21 formalism that the evaluation ignores. Reviewers reward a model whose theorem changes what you22 build or bound.23- **Pair every claim with proportional evidence** — a proof for an analytical claim, a24 simulation/testbed number with the setup named, or a measurement with its methodology — not25 adjectives like "significantly improves."26- **Write defensively.** Name the assumption a theory reviewer will question and justify it; name27 the baseline a systems reviewer will call unfair and tune it; state the regime where your result28 does *not* hold. Because there is no rebuttal, an unaddressed objection becomes a reject.29- **Respect the page budget as a design constraint.** Nine pages of text (appendices included) is30 tight; a proof sketch in the body with the full proof compressed, and a focused evaluation, beat31 an over-scoped paper that only fits by cutting the model.32- **Maintain double-blind wording** in self-citations (third person), tool names, testbed/cluster33 identifiers, acknowledgements, and funding.3435## Networking-paper skeleton3637| Section | Job it must do | Common failure |38|---|---|---|39| Intro | Problem, setting, gap, contribution, what improves — by the first column or two | Leads with a trend, not a networking problem |40| System model | Topology, traffic, constraints, assumptions, notation, stated plainly | Assumptions hidden or introduced only in the evaluation |41| Analysis / design | Theorems + proof sketches, or the protocol/algorithm, reproducibly | A model no result uses; a design too thin to re-implement |42| Evaluation | Each claim answered with proportional evidence, honest baselines | Toy topology; untuned baseline; metric that proxies the claim |43| Discussion / limits | The regime where the result holds; deployment caveats | No limits stated (invites the objection you cannot rebut) |44| Related work | Delta-first positioning against the networking literature | Citation catalog with no contrast |4546## Sentence-level rewrites4748| Draft pattern | INFOCOM-safe rewrite |49|---|---|50| "Our scheme significantly improves throughput." | "improves median throughput by X% (95% CI ...) over <tuned baseline> in <topology/scale>" |51| "We prove our algorithm is optimal." | "We prove a (1-1/e) approximation under <stated assumptions>; §V bounds the gap empirically" |52| "We evaluate on a large simulation." | "We simulate <N nodes> in <ns-3/custom>, <traffic model>, <M seeds>; setup in §V-A" |53| "The model captures the system." | "Under assumptions A1-A3 (stated and justified in §III), the model yields Theorem 1" |54| "State-of-the-art performance." | Claim scoped to the topology, traffic, and regime actually tested |5556## Defensive-writing discipline (no rebuttal)5758```text59[Assumption] which modeling assumption will a theory reviewer challenge? -> justify it in §III60[Baseline] which comparison will a systems reviewer call unfair? -> tune it, document the budget61[Scale] does the evaluation reach a realistic size? -> if not, scope the claim and say so62[Regime] where does the result break? -> state it before a reviewer finds it63-> each objection answered in the PDF, because there is no turn to answer it in a reply64```6566## Vignette: compressing an over-scoped analytical paper6768A draft with two models, five theorems, and a sprawling evaluation: keep the one model the design69uses, the two theorems that carry the insight (full proofs compressed or sketched), and the two70evaluation figures that test the headline claim; move notation-heavy derivations into a tight71in-budget appendix (remember it counts toward the nine pages); cut the second model to a remark.72The test of a good cut: a reviewer should be able to state your assumptions, your main result, and73why your baseline is fair — from the body alone.7475## Output format7677```text78[Writing diagnosis] clear / under-modeled / over-claimed / evidence-mismatched / over-scoped79[First-page fix] <new framing leading with the networking problem + system model>80[Claim audit] <claim -> proof or simulation/measurement -> proportional? yes/no>81[Defensive gaps] <assumption/baseline/scale/regime objection -> where to pre-empt it>82[Anonymity edits] <self-citations / tool / testbed names / acks to rewrite>83```
brycewang-stanford