eurosys-writing-style

EuroSys Writing Style

Use this on the manuscript itself. A EuroSys paper succeeds when a broad

systems PC — kernel people, distributed-systems people, cloud and security

people — can each extract the same mechanism and believe the same numbers

within one reading. The two-column acmart[sigplan] format and the 12-page

technical budget (EuroSys 2027 CFP, rendered 2026-07-08) reward density that

stays legible, not compression that hides.

The load order of a EuroSys introduction

1. The operational pain, stated concretely enough to measure.
2. Why the existing abstraction or mechanism structurally cannot fix it —

named systems, not "prior approaches".

1. The insight: the one idea that unlocks the design.
2. What was built, and the two or three numbers that prove the idea carries.
3. Contributions as verifiable statements, each pointing at a section.

A reader who stops after page one should be able to state your insight in a

sentence. If the insight sentence does not exist, the problem is the research

framing, not the writing (route to eurosys-topic-selection).

Page budget that survives review

| Section | Share of 12 pages | Failure mode when misallocated |

|---|---|---|

| Intro + motivation | ~2 | Motivation essay crowds out the design |

| Design + implementation | ~4.5 | Mechanism described by metaphor, not by structure |

| Evaluation | ~4 | Result dump with no questions being answered |

| Related work + discussion | ~1.5 | Either a stub or a survey; both read badly |

References are free pages at EuroSys — cite generously and spend zero

technical pages on bibliography anxiety.

Design sections that reviewers can retell

• Name components and keep the names stable; a renamed component is a new

component to a tired reviewer.

• One architecture figure whose boxes match the subsection structure.
• For each mechanism: the invariant it maintains, the common-case path, the

failure path, and the cost you pay for it. The cost sentence is what

separates engineering description from research writing.

• Push code-level detail down to a listing only when the trick is the code.

Evaluation prose discipline

Open with the questions, then answer them in order:

§6 Evaluation. We answer four questions:
  Q1  Does <system> improve end-to-end <metric> on realistic workloads? (§6.2)
  Q2  Where does the gain come from? (§6.3, component breakdown)
  Q3  What does it cost — memory, CPU, complexity? (§6.4)
  Q4  When does it lose? (§6.5, adversarial and edge workloads)

Q4 is the EuroSys credibility section: a paper that measures its own worst

case defuses the reviewer whose job is to find it.

• Every figure caption states the takeaway, not just the axes.
• Numbers in prose carry their conditions: "2.1x on the write-heavy trace at

p99", never bare "2.1x better".

• Banned without qualification: "significantly", "dramatically", "orders of

magnitude", and any superlative the data does not literally show.

Sentence surgery

| Draft reflex | EuroSys register |

|---|---|

| "X is a critical challenge in modern datacenters" | Open with the measured symptom, not the genre claim |

| "significantly reduces latency" | "reduces p99 by 38% on trace T at 85% load" |

| "our novel approach" | Name the mechanism; novelty is the reviewer's verdict |

| "due to space constraints, details are omitted" | Point to the artifact or cut the claim |

| "we believe this generalizes" | State the condition under which it generalizes, or drop it |

| "as shown in Figure 6" (alone) | "...which isolates the planner's contribution (Fig. 6)" |

Revision passes that pay at this venue

1. Retell pass — after each design subsection, write its three-sentence

PC-meeting summary; if you cannot, the subsection is not done.

1. Number pass — every performance number in prose gets its conditions

attached or gets deleted; captions get takeaways.

1. Skeptic pass — read only §Evaluation as a hostile reviewer hunting

untuned baselines and missing worst cases; fix what you find.

1. Blind pass — search for lab names, grant strings, first-person

references to prior systems, and telltale cluster hostnames in listings.

1. Budget pass — measure section lengths against the 12-page allocation

and rebalance before polishing sentences that will be cut anyway.

Double-blind voice

• Your prior systems appear in third person: "extends Foo [12]" — even when

the extension is obviously yours.

• Deployment anecdotes get de-identified without being falsified: "a European

cloud operator" is fine; invented deployments are fraud.

• Acknowledgments, grant numbers, and lab-specific artifact URLs wait for

camera-ready.

Output format

[Insight sentence] <the one-sentence idea a reviewer should retell>
[Arc audit] pain -> structural gap -> insight -> system -> numbers: present/missing
[Budget skew] <sections over/under the 12-page allocation>
[Claim hygiene] <unscoped numbers and superlatives found>
[Q4 status] <does the evaluation measure where the system loses?>