icde-review-process

ICDE Review Process

Use this to reason about review-stage strategy. Reopen the current call, important-dates page,

and any reviewer instructions before making process claims; ICDE re-declares its review model

each edition.

Process model

• ICDE takes research papers in two rounds per edition (ICDE 2027: ≈June and November).

Each paper draws at least three reviewers, coordinated by an area chair who

synthesizes the reviews into a recommendation.

• The ICDE 2026 model ran two reviewing phases inside a round: the first phase returned

Revise & Resubmit, Accept, or Reject; invited revisions were re-reviewed (≈4-week window)

before a final Accept/Reject. Whether the current edition keeps the R&R phase is

edition-specific (待核实).

• Review is single-blind: reviewers see author identity; authors do not see reviewers. This

changes the calculus — reputational priors exist, so the paper must earn belief on evidence,

not on names.

• A Round 1 reject is terminal for the edition; the paper cannot re-enter Round 2. That

raises the stakes on getting the evaluation right the first time.

• Accepted papers publish in IEEE Xplore; camera-ready compliance and IEEE metadata matter

as much as the acceptance itself.

Who reviews here

• The committee is builder-heavy: reviewers who have implemented indexes, optimizers, and

storage engines. They read an evaluation the way an engineer reads a benchmark — skeptical of

untuned baselines, single-point results, and hidden costs.

• Because scope overlaps SIGMOD and VLDB, at least one reviewer will likely know the nearest

prior system intimately; a novelty claim that ignores it is caught, not skimmed past.

• Borderline systems papers usually fail on one of three edges: an untuned or unfair baseline, a

performance claim with no operating point, or a mechanism whose benefit is never isolated by

an ablation.

Scoring leverage table

| Review dimension | What raises it | What sinks it |

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

| Novelty | A named mechanism a builder could not have trivially derived | "Yet another index" with no isolable new idea |

| Evaluation | Tuned baselines, scale curves, variance, disclosed cost | Single-point wins on one synthetic workload |

| Soundness | Ablations that isolate the mechanism's contribution | Gains that could come from unrelated engineering |

| Reproducibility | Runnable supplement whose availability is evident | Empty or unrunnable supplemental material |

| Significance | A primitive an ecosystem can adopt | A narrow tweak with no downstream reach |

Stage-by-stage realism

• Phase 1 reviews: read them as the area chair will — is the central evaluation objection

answerable? A Revise & Resubmit is an opportunity, not a soft reject; treat the requirement

list as binding.

• Revision re-review: the same reviewers check the list; unrequested scope changes hurt.
• Decision: one unresolved fairness or cost objection outweighs several resolved clarity

nits, because a builder-heavy committee anchors on the evaluation.

• Round choice as strategy: since Round 1 rejects skip the edition, aiming an immature

evaluation at the earlier round is a common self-inflicted loss.

Output format

[Current stage] submitted / phase-1 reviews / revision / decision / camera-ready
[Round] Round 1 or Round 2 (edition)
[Decision actors] <reviewers / area chair / program chairs>
[Likely leverage] <novelty / evaluation / soundness / reproducibility>
[Central objection] <the one thing the area chair will weigh most>
[Next response move] <one action>