percom-author-response
GitHub用于撰写 IEEE PerCom 单轮反驳信，针对审稿人疑问提供基于原文证据的直接回答，纠正误解并澄清关键点。严格遵循双盲原则，不承诺新实验，确保每个问题都有据可查，以争取录用。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill percom-author-response -g -y
SKILL.md
Frontmatter
{
"name": "percom-author-response",
"description": "Use when drafting an IEEE PerCom rebuttal after being invited following the initial reviews, covering the bounded single-round reply that answers explicit reviewer questions, corrects misreadings, points to evidence already in the paper without promising new experiments, and preserves double-blindness."
}
PerCom Author Response
Use this after PerCom reviews are released and you have been invited to submit a rebuttal. PerCom's rebuttal is a single, bounded turn: it answers the reviewers' explicit questions, resolves misunderstandings, and clarifies — and the call states that new experiments are not expected. There is no journal-style revision that follows, so the rebuttal must win with the paper you already submitted. It remains double-blind: reveal no author, institution, testbed, or dataset-owner identity.
First: did you clear the early-rejection gate?
Papers with no positive review are early-rejected before the rebuttal opens — there is no rebuttal to write for those. If you were invited, at least one reviewer sees a path to publication: your job is to give that advocate the answers they need for the TPC discussion, and to neutralize the specific doubts of the others.
Triage
- Answer what affects the decision: significance of the contribution, soundness of the method, cross-subject / deployment evaluation, appropriate metrics (F1 vs. pooled accuracy), and clarity.
- Use evidence that already exists in the submitted paper or dataset — a table, a figure, an appendix result the reviewer overlooked. Do not promise a new experiment; the process does not expect or reward one, and you cannot add it.
- Correct factual misreadings first; a reviewer who misread which split produced a number is often persuadable with a pointer.
- Keep every word anonymous — do not name your testbed, deployment site, institution, or a non-anonymized dataset, even to strengthen a point.
Structure the bounded reply
Group by reviewer, lead each item with the reviewer's own question, and answer with a pointer into the existing paper:
[R1.Q1] "Is the eating-recognition result cross-subject or within-subject?"
-> Table 2 already reports leave-one-subject-out F1 (0.xx, 95% CI ...); the pooled number
in the abstract is secondary. We will make this ordering explicit in §4.1.
[R2.Q1] "The baseline seems untuned."
-> The baseline used the same feature budget and a grid search reported in Appendix A;
we point R2 to Table A1 and will surface it in the body.
[R3.Q1] "How does it behave under confounding non-eating activities?"
-> Table 4 already measures false positives against typing/brushing; we clarify the caption.
The rule that wins a bounded rebuttal: every explicit question gets a direct, located answer. A question left unaddressed is what the discussion punishes, because there is no later round to fix it.
Reviewer pushback patterns
| Pushback | What it signals | PerCom-ready response |
|---|---|---|
| "Evaluation is within-subject" | The core ubicomp doubt | Point to your leave-one-subject-out table; if you truly lack it, this is likely fatal — do not fake it |
| "Accuracy is inflated by class imbalance" | Metric-choice doubt | Point to reported F1 / event-level metrics and the class balance; clarify which metric is headline |
| "The baseline is not fair/tuned" | Soundness doubt | Point to the equal-budget tuning already in the paper/appendix; clarify placement |
| "Does this generalize beyond your subjects?" | External-validity limit | Point to the diversity of subjects and the stated limitation; do not overclaim |
| "The dataset/system is not available" | Open-data gap | Point to the (anonymized) release plan; describe what will be public post-acceptance |
| "Contribution overlaps prior work X" | Novelty doubt | Sharpen the delta in words; point to the comparison already present |
Anonymity in the rebuttal (easy to slip)
- Refer to your own prior work in the third person, as in the paper.
- Describe the dataset/testbed without naming the site, building, or repository owner; use the anonymized location.
- Do not thank a named collaborator, funder, or institution inside the reply.
Calibration
- Respond to the criterion the reviewer actually raised, not the one you would rather defend.
- The rebuttal length limit and format vary by cycle; confirm the current instructions and stay inside the limit — over-length rebuttals read as unfocused.
- The TPC discussion decides; write the reply as ammunition for your advocate, and let the paper — not the rebuttal — carry the argument.
Output format
[Gate] invited to rebuttal? yes/no (if no: reroute, do not draft)
[Priority question] <reviewer's explicit question>
[Decision dimension] significance / soundness / cross-subject evaluation / metrics / clarity
[Located answer] <question -> pointer to existing table/figure/appendix; clarification to add>
[No-new-experiment check] <every answer grounded in submitted material? yes/no>
[Anonymity check] <no identity leak in the reply: passed/issues>
Version History
- 9f86f09 Current 2026-07-19 17:19
