isca-author-response
GitHub指导ISCA论文作者在三周反驳期内高效处理审稿意见。通过分类异议（事实错误、证据缺口等）并评估杠杆率，制定实验计划与回复策略，旨在为PC会议冠军提供有力支持，避免无效争论。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill isca-author-response -g -y
SKILL.md
Frontmatter
{
"name": "isca-author-response",
"description": "Use when ISCA reviews arrive and the rebuttal-plus-revision window opens — triaging objections by what a three-week window can actually fix, deciding between textual rebuttal and a revised PDF with new experiments, structuring responses that arm a champion at the PC meeting, and avoiding window-burning mistakes."
}
ISCA Author Response
ISCA 2026 gave authors a combined rebuttal/revision period, February 16 to March 6 — three weeks in which the team could both argue and change the paper. That design rewards teams that treat the window as an engineering sprint with a triage phase, and punishes teams that spend it composing indignation. Everything here assumes that verified 2026 structure; the 2027 window's existence, length, and rules are 待核实 until the new CFP posts.
Day 1-2: triage before any writing or computing
Read all reviews twice: once fast for temperature, once slow building the objection ledger. Classify every distinct point:
| Class | Definition | Window response |
|---|---|---|
| F — factual error | Reviewer misread something present in the PDF | Quote page/section/figure; one sentence; zero new work |
| E — evidence gap, closable | Missing experiment your infrastructure can run in days | Run it; report in response; fold into revision if permitted |
| E∞ — evidence gap, not closable | Needs silicon, months, or data you lack | Concede scope honestly; state what the existing evidence does cover |
| D — design objection | "This breaks under X" / "cost is understated" | Analyze X concretely; if real, add the limitation; if not, show why |
| P — premise objection | "This problem doesn't matter" | Strongest countermeasure is motivation data; if you have none beyond the paper, one calm paragraph and move on |
| T — taste | "I'd have designed it differently" | Acknowledge once; do not redesign the paper |
Then rank by leverage: an F held by two reviewers outranks an E held by one;
anything a likely champion needs to defend the paper outranks everything else
(isca-review-process explains why the champion is the audience).
Day 2: freeze the experiment plan
Pick at most three new experiments — the ones that convert the highest-leverage E-class objections into numbers. Assign owners and a hard finish date one week before window close, leaving prose time. The classic window failure is five half-finished experiments and a response that promises rather than shows.
window-plan.md
R2.O3 (E, leverage=high) "no comparison vs LazyMerge at equal storage"
-> run: lazymerge @ 18KB/core, full suite; owner: MK; due Feb 27
R1.O1 (F, leverage=high) "warm-up unstated"
-> answer: §5.1 para 2, 50M inst; also add to config table in revision
R3.O2 (D, leverage=med) "decay threshold fragile"
-> run: sweep T in {2K..64K}; expect plateau; owner: JL; due Feb 25
R3.O5 (T, leverage=low) "why not epoch-based?"
-> one paragraph, cite §4.2 rationale; no experiment
Writing the response document
Structure for committee reading, not reviewer combat:
- Opening block (5-6 lines): thank, then list the concrete changes/results produced during the window — the things a champion can repeat in the room.
- Per-objection sections keyed to review IDs, highest-leverage first. Each: restate the objection in one neutral line, answer with evidence (number, table row, or PDF pointer), stop.
- Shared-objection consolidation: answer once, cross-reference the copies — it shows the committee the objection set is smaller than it looks.
Tone rules that survive contact with real committees: concede real limitations in plain declarative sentences (concessions buy credibility that transfers to your rebuttals); never speculate about reviewer expertise or identity; never claim "all reviewers agree" about anything they don't.
Using the revision half of the window
If the cycle permits an updated PDF (2026's window was named rebuttal/revision):
- Make surgical edits keyed to objections: the missing baseline column, the config-table row, the limitation paragraph, the clarified walk-through.
- Do not restructure the paper — reviewers re-reading diff mentally against their memory, and a rewritten paper resets their evaluation instead of repairing it.
- Maintain a change log mapping edit → objection; if the process provides no place to submit it, embed the mapping in the response document.
- Keep the paper inside all format rules; a revision that breaks the 11-page
budget converts goodwill into a violation. This is what the half-page of slack
from
isca-writing-stylewas for.
What not to spend the window on
- New contributions. Results beyond the paper's claimed scope read as a different paper and invite "needs another round of review" — the phrase that kills borderline papers.
- Rebutting every sentence. A 15-point response to a 3-objection review buries your leverage. Committees skim; make the skim land the top three.
- Premise litigation at length. One paragraph of motivation defense with data; more looks defensive and reads worse each repetition.
- Simulator relitigating. If a reviewer distrusts your instrument globally, the fix is the validation anchor you either have (point to it) or don't (concede and scope) — not an essay on simulation philosophy.
Window-close checklist
- Every review ID answered or explicitly consolidated
- Top-3 objections answered with numbers produced this window
- All F-class errors corrected with pointers, tone neutral
- Revision (if permitted) diff-minimal, format-legal, change-mapped
- A colleague outside the project read the response cold and could state your three key answers back to you
- Submitted with margin — the window closes AoE like everything else
The three weeks on a calendar
Using the 2026 window (Feb 16 - Mar 6) as the shape:
| Days | Activity |
|---|---|
| 1-2 | Triage; objection ledger; leverage ranking; no computing yet |
| 2 | Experiment plan frozen: ≤3 runs, owners, due dates |
| 3-12 | Experiments execute; response skeleton drafted in parallel |
| 13-15 | Results in; response written against real numbers |
| 16-18 | Revision edits (if permitted), diff-minimal, format-checked |
| 19-20 | Cold read by an outside colleague; tone pass; consolidation pass |
| 21 | Submit with hours to spare, AoE-checked |
After the window, silence until decisions; outcome handling branches in
isca-review-process, and calendar context lives in isca-workflow. Verified
dates: ../../resources/official-source-map.md (2026-07-08).
Version History
- 9f86f09 Current 2026-07-19 16:05
