isca-review-process

name	isca-review-process
description	Use when reasoning about how an ISCA submission is evaluated — the two-round review structure with December and February reviewer waves, what the combined rebuttal-and-revision window means for decisions, PC-meeting dynamics at a flagship architecture venue, and how to read the eventual outcome.

ISCA Review Process

Knowing the pipeline changes author behavior at three points: how the paper is written (for two waves of readers), how the February window is used (it permits revision, not just argument), and how the decision is interpreted. The stage structure below is the verified ISCA 2026 calendar; internal committee mechanics are described as community-typical practice, labeled as such, because programs rarely publish them.

The 2026 pipeline as verified

Stage	2026 dates	Author-visible?
Abstract registration	Nov 10, 2025	yes — HotCRP
Paper submission	Nov 17, 2025	yes
Round 1 reviews written	due Dec 19, 2025	no (delivered later)
Round 2 reviews written	due Feb 13, 2026	no
Rebuttal + revision window	Feb 16 - Mar 6, 2026	yes — the pack's pivot point
PC deliberation and decisions	after the window; exact 2026 date 待核实	outcome only
Conference	Jun 27 - Jul 1, 2026, Raleigh	yes

What two rounds imply

A two-round structure typically means an early wave of reviews on every paper, after which papers that remain viable receive additional expert reviews in the second round. Two operational consequences for authors:

• The paper must survive a first reading by a possibly non-specialist architect. Round 1 breadth is why the introduction and motivation section carry so much weight (isca-writing-style): a paper that requires subarea fluency to appreciate may not reach the readers who have it.
• Review counts differ across papers. Receiving five or six reviews usually signals second-round attention; receiving few does not reliably signal anything. Do not decode fate from counts — the window that follows is the same either way.

The combined rebuttal/revision window

The distinctive verified fact of the 2026 cycle: authors got a three-week window (Feb 16 - Mar 6) described as a rebuttal/revision period — long enough to run experiments and change the PDF, not merely to file 500 words of protest. Venues design such windows so the committee can conditionally like a paper: reviewers name concrete deficiencies, authors demonstrate the fix is real, and the PC judges the demonstrated fix rather than a promise. Strategy and mechanics for the author side live in isca-author-response; the process-level points:

• Reviews arriving at window-open are the complete input you will get; there is no later dialogue round to clarify misreadings. Everything must land in this window's artifacts.
• Reviewers reconvene (online discussion, then a PC meeting) with your response and any revision in front of them. The audience is the whole paper's discussion group, including reviewers who liked it and need ammunition to defend it.

Inside the committee room (typical practice, not CFP text)

Flagship architecture PCs conventionally work discussion lists ordered by average score, with each paper championed or buried in minutes of conversation. Patterns worth internalizing:

• A paper without a champion — one reviewer willing to spend social capital — rarely survives discussion regardless of decent scores. Rebuttals should be written to arm the champion, not to defeat the detractor.
• Expertise asymmetry is normal: the most negative review is often the least expert. The committee usually knows this; a calm, factual response lets them discount it. An irritated response makes the room defend its colleague.
• Methodology objections travel well in the room ("the baseline is untuned" fits in one sentence); nuanced defenses do not. This is why methodology rigor up front (isca-experiments) is cheaper than any rebuttal.
• Conflicted members leave the room; your conflict declarations are what make that machinery work (isca-submission).

Reading the outcome

ACCEPT            -> isca-camera-ready + isca-artifact-evaluation immediately;
                     June presentation logistics.
ACCEPT-with-shepherd (if used) -> the shepherd's list is a contract; satisfy it
                     completely and visibly in the camera-ready.
REJECT, reviews engage the mechanism -> the idea was heard; fix the named
                     deficiencies and retarget (MICRO spring / HPCA summer)
                     with the reviews as a to-do list.
REJECT, reviews dispute the premise -> the community doesn't buy the problem;
                     more results won't help — reframe or re-venue
                     (isca-topic-selection).
REJECT, reviews are thin/confused   -> often a routing casualty; check whether
                     your abstract-step topics steered assignment badly before
                     blaming the idea.

Whether ISCA 2027 uses the same round structure, offers a revision phase at all, or publishes different outcome categories is unverified — the 2027 CFP was not posted at check time (待核实).

Confidentiality and conduct

Submissions are confidential review material: do not publicize reviews, do not contact suspected reviewers, and route irregularities (leaked identity, hostile review, suspected conflict violation) to the program chairs through official channels only. Public deadline-week complaints are read by exactly the community that reviews your next five papers.

Author-side calendar hooks

• At submission: file the five hardest anticipated objections with pre-computed answers (isca-workflow makes this a named role).
• Window-open minus one week: environment resurrection drill — the simulator tree, configs, and workloads must run (isca-reproducibility).
• Window-open: triage day (see isca-author-response); no experiments before triage completes.
• Decision day: run the outcome tree above within 48 hours while morale allows planning.

Calendar facts from ../../resources/official-source-map.md (verified 2026-07-08); committee-mechanics paragraphs are labeled community practice and should be re-weighed against any process description the 2027 chairs publish. Whenever a live 2027 reviewer-instructions page appears, it supersedes every "typical practice" paragraph in this file — update the source map first, then this skill.