Series ID: EDKS-ADV-ZH-0023 · How to Learn Advanced English (Chinese Edition) · Lesson No.023 · C1 → C2

Revise a Research Paper After Peer Review Without Breaking Its Logic | 根据 Peer Review 修订论文，但不要把整篇逻辑修坏

Peer review does not merely ask you to “fix comments.” A serious revision changes a living argument. Every new analysis, clarification, deleted claim, added limitation or revised title can alter other sections of the manuscript. If you treat reviewer comments as isolated editing tickets, you can improve twenty sentences and quietly break the logic of the paper.

Peer review 后的 revision 不是“逐条改 comment”这么简单。每一个 reviewer request 都可能改变 research question、analysis hierarchy、claim strength、scope、Title、Abstract、Discussion 或 Conclusion。修稿真正难的地方，是让 paper 变得更好，同时让整篇 research identity 仍然是一体的。

This lesson therefore treats revise-and-resubmit as scientific change control.

The governing sequence is:

DECISION LETTER → CLASSIFY REQUEST → DECIDE RESPONSE → MAKE SCIENTIFIC CHANGE → PROPAGATE DEPENDENCIES → DOCUMENT CHANGE → RE-AUDIT WHOLE PAPER → RESPOND POINT BY POINT.

Decision letter → 分类 request → 决定 response → 修改 science → 传播 dependent changes → 记录 change → 重审整篇 paper → point-by-point 回复。

---

Part I — Revision is change control, not comment chasing | 第一部分：修稿是 change control，不是 chasing comments

1. Start with the editor, not Reviewer 1 | Editor owns the decision

The editor’s decision letter frames the revision. Reviewers provide expert advice; the editor decides what must be resolved for the journal to reconsider the manuscript.

Before touching the paper, extract:

• decision type;
• deadline;
• editor-priority concerns;
• required experiments/analyses;
• formatting or policy requests;
• whether the revision is likely to return to reviewers;
• submission-file requirements.

2. Nature: revised manuscripts commonly return to reviewers | Your response letter is part of the scientific conversation

Nature’s reviewer guidance notes that reviewers are normally asked to assess revisions and receive the revised manuscript and response context. That means the rebuttal is not a private emotional note; it is a working document written for scientific reassessment.

Nature | Revisions and Feedback

3. Nature Computational Science 2025: start with a concise revision summary | Then go point by point

A 2025 Nature Computational Science editorial recommends beginning the response document with a short summary of the main concerns and changes, followed by complete point-by-point responses. It also recommends preserving the reviewer’s full comment rather than selectively quoting only the easiest part.

Nature Computational Science | How to Respond to Reviewers

4. Springer Nature: address every point, but disagreement is allowed | Compliance is not obedience

Springer Nature’s revision guidance asks authors to address all reviewer/editor points, describe major revisions, undertake appropriate new work, and provide a scholarly explanation when a suggested change is not made because the authors believe it would not improve the paper.

Springer Nature | Revising and Responding

5. Elsevier: responses should be polite, objective and evidence-based | Disagreement can be professional

Elsevier similarly states that authors may disagree with reviewer suggestions when they explain the reasoning clearly and respectfully, and that reviewer misunderstandings may indicate the manuscript itself needs clearer explanation.

Elsevier | Peer Review and Resubmission

6. The first rule of revision | Do not edit while emotionally reacting

A harsh or confusing report can make authors defend sentences before understanding the underlying problem. Separate reaction from diagnosis.

Read the decision letter once for emotion.

Read it again for tasks.

Read it a third time for manuscript dependencies.

7. Reviewer tone and scientific content are separate variables

A rude comment can contain a valid scientific concern.

A polite comment can contain a factual error.

Your response should address the scientific content.

8. Nature Geoscience: keep rebuttals dispassionate | Raise serious concerns with the editor separately

Nature Geoscience advises authors to keep responses constructive and reviewer-facing, while concerns about inappropriate review behaviour can be raised separately and specifically with the editor.

Nature Geoscience | The Art of Responding to Reviews

9. Build a revision control sheet before making any change | One row per request

ID	Source	Request	Type	Decision	Manuscript dependencies	Status
E1	Editor	clarify primary claim	scientific framing	accept	Title, Abstract, Discussion, Conclusion	open
R1.2	Reviewer 1	add subgroup analysis	new analysis	consider	Methods, Results, Discussion, response	open
R2.4	Reviewer 2	replace outcome terminology	construct issue	partly accept	lexicon, all sections	open

10. Give every comment an ID | R1.1, R1.2, R2.1…

Without IDs, long revisions become difficult to track. IDs also help co-authors discuss one request without ambiguity.

11. Separate editor comments from reviewer comments | Priority matters

Use:

E1, E2…

R1.1, R1.2…

R2.1, R2.2…

12. Do not begin by assigning comments to co-authors | Classify first

If you immediately send “statistics comments to statistician” and “writing comments to first author,” you can miss cross-section effects.

13. Revision request taxonomy | Eight useful classes

1. Scientific validity concern.
2. Additional analysis request.
3. Additional experiment/data request.
4. Interpretation/scope concern.
5. Clarity/structure concern.
6. Literature/context request.
7. Policy/format/reporting request.
8. Reviewer misunderstanding or factual error.

14. Scientific validity concern | Highest priority

Examples:

• wrong analysis;
• confounded design;
• invalid outcome interpretation;
• missing critical control;
• data inconsistency;
• unsupported causal claim.

These can change the paper’s research identity.

15. Additional analysis request | Not all analyses are equal

Ask:

• Does the analysis answer the original question?
• Is it exploratory?
• Is sample size adequate?
• Does it require a new assumption?
• Would it change primary conclusions?

16. Additional experiment request | New data can create a new paper identity

A reviewer may ask for an experiment that materially expands the scope. Before agreeing, determine whether the revision remains the same study or becomes a new research programme.

17. Interpretation/scope concern | Often solved by narrowing, not more data

Example:

Reviewer: “The manuscript claims durable learning, but follow-up ends at four weeks.”

Possible best response:

change the claim from “durable” to “four-week persistence.”

18. Clarity concern | Misunderstanding is evidence about the manuscript

If an expert reader misunderstands a central point, do not merely reply “the reviewer misunderstood.” Ask what sentence, definition or figure allowed that misunderstanding.

19. Literature request | Add only if it changes context or fairness

Do not create a citation dumping ground to satisfy every suggested paper. Add relevant work and explain how it changes framing.

20. Reporting/policy request | Usually non-negotiable if journal requires it

Examples:

• data availability;
• ethics statement;
• CONSORT/STROBE/PRISMA checklist;
• reporting of sample size;
• figure format;
• conflict-of-interest disclosure.

21. Reviewer factual error | Correct politely and precisely

Useful response structure:

We thank the reviewer for raising this point. We believe the concern arises from X. In the submitted manuscript, Y was defined as… To make this distinction clearer, we have now revised lines…

22. Reviewer asks for an impossible analysis | Explain why it is invalid, not merely inconvenient

Weak:

We did not do this because it would take too much time.

Stronger:

We did not fit the requested model because the subgroup contains only six outcome events, which would make the estimate unstable and potentially misleading. We instead report descriptive estimates and identify the analysis as a future confirmatory question.

23. Reviewer asks for analysis outside the research question | Decide whether it adds value

You may:

• perform it as clearly labelled exploratory work;
• decline with rationale;
• provide it in supplement if useful but not central.

24. Reviewer asks for a new primary outcome after seeing results | Preserve temporal history

Do not rewrite the paper so the new analysis appears prespecified.

25. Reviewer asks to “strengthen the conclusion” | Strength must come from evidence, not wording

Sometimes the correct revision is to clarify the contribution, not increase certainty.

26. Reviewer asks to “tone down” claims | Check whether the design actually supports the original claim

Do not automatically weaken a justified result. Explain the design if the reviewer has underestimated it.

27. Reviewer disagreement matrix | Accept / Partly accept / Disagree / Cannot address

Decision	Meaning
Accept	change improves manuscript and is valid
Partly accept	underlying concern valid; requested solution modified
Disagree	request would be scientifically incorrect or unnecessary
Cannot address	data/resources/design do not permit request; explain consequences

28. “Partly accept” is underused | Separate concern from proposed remedy

Reviewer:

The analysis does not address baseline imbalance; use propensity-score matching.

You may agree that confounding needs stronger treatment but choose regression adjustment or weighting for defensible reasons.

29. Do not disagree with the reviewer’s solution when you agree with the problem | Solve the problem

30. Major revision vs minor revision | Scientific meaning, not emotional meaning

A “major revision” may involve significant additional analyses or conceptual restructuring. “Minor” can still require careful changes. Treat both as controlled manuscript changes.

31. Decision-letter hierarchy | Editor request > reviewer preference when conflict exists

If reviewer requests conflict, the editor’s framing can guide priority. If conflict materially affects the revision, explain the trade-off in the response and, where appropriate, ask the editor for clarification.

32. Conflicting reviewers | Do not satisfy both by making the paper incoherent

Reviewer 1:

Shorten the Discussion.

Reviewer 2:

Add more limitations and mechanism discussion.

Solution:

remove repeated Results, then use the saved space for higher-value limitation/mechanism analysis.

33. Conflicting statistical requests | Choose the scientifically justified model

Do not run one primary model for Reviewer 1 and a contradictory primary model for Reviewer 2. State which analysis is primary and why; use alternatives as sensitivity analyses if appropriate.

34. Reviewers are not co-authors | Authors retain responsibility

You are responsible for the final scientific validity of the paper. “Reviewer asked us to” is not a defence for an invalid analysis.

35. The revision map | Every accepted change needs dependency analysis

For each accepted comment, ask:

If we change this, what else must change?

36. Dependency categories

• Title;
• Abstract;
• Introduction;
• Methods;
• Results;
• Figures/tables;
• Discussion;
• Conclusion;
• Supplement;
• Response letter;
• protocol/registration statement if relevant.

37. Example: new primary analysis changes the estimate

Reviewer requests clustered standard errors; authors agree.

New uncertainty interval changes.

Dependencies:

• Methods statistics;
• Results estimate;
• Abstract result;
• figure confidence intervals;
• Discussion certainty;
• Conclusion certainty.

38. Example: terminology correction

Reviewer notes that “competence” is too broad; scale measures self-efficacy.

Dependencies:

• Title if term appears;
• Abstract;
• Introduction construct definition;
• Methods instrument description;
• Results labels;
• Discussion interpretation;
• Conclusion.

39. Example: limitation added

Reviewer identifies single-site selection concern.

If accepted, do not add only one sentence to Limitations. Recheck:

• Abstract conclusion;
• Discussion generalisation;
• Conclusion;
• Title if broad population claim exists.

40. Example: exploratory subgroup added

Reviewer suggests baseline-proficiency subgroup analysis.

Dependencies:

• Methods label as post hoc/exploratory;
• Results subgroup estimate;
• Discussion cautious interpretation;
• no rewriting of Introduction as prespecified hypothesis;
• possibly Supplement if secondary.

41. Never hide revision chronology | Revision should not rewrite history

Scientific manuscripts can change. The key is to preserve when decisions occurred.

42. “We added this analysis in response to Reviewer 2” is useful information

When appropriate, state in the response letter that the analysis was added during revision. In the manuscript, label it exploratory/post hoc where that status affects interpretation.

43. New analysis can become important without becoming prespecified

Exploratory does not mean unimportant. It means the evidence entered the reasoning at a later time.

44. Revision should improve temporal honesty | Not erase it

45. Point-by-point response structure | Make assessment easy

For every comment:

1. Reviewer comment.
2. Author response.
3. Action taken.
4. Location in manuscript.
5. Quoted revised text where useful.

46. Nature Computational Science: preserve full reviewer comments | Avoid selective quotation

This helps editors/reviewers see that every concern has been addressed in context.

47. Response letter should be independently readable | Do not write “done”

Weak:

Done.

Weak:

Fixed in manuscript.

Strong:

We agree that the original Methods did not make the exclusion rule sufficiently explicit. We now state that participants missing the one-week primary outcome were excluded from the complete-case primary analysis and report the resulting n in both Methods and Results (Methods, lines 212–219; Results, lines 331–336).

48. A good response explains the scientific change, not just the text change

Changed “effective” to “associated with” is useful.

We revised the causal language because the non-randomised design does not identify a treatment effect is better.

49. Response summary at the top | Help editor understand the revision globally

Useful summary:

• primary claim narrowed;
• new sensitivity analysis added;
• sample-flow inconsistency corrected;
• mechanism claim removed;
• limitations expanded;
• figures revised.

50. The response document has its own coherence | Main changes should match detailed responses

If the summary says “no conclusions changed” but detailed responses show the primary claim was narrowed, the response document itself is inconsistent.

51. Track changes and clean manuscript | Different purposes

The marked manuscript helps reviewers locate revisions.

The clean manuscript helps editors read the final paper as a paper.

Elsevier guidance commonly recommends providing both when requested.

Elsevier | Responding to Reviewer Comments

52. Line numbers make revision verifiable | Especially for long manuscripts

Use stable line numbering in the revised file if the journal requests or benefits from it.

53. Quoting revised text can reduce reviewer effort

Especially useful when the change is short and central.

54. Do not quote five pages of new Methods in the response | Summarise and point to location

55. Respectful disagreement template | Evidence before rhetoric

We appreciate the reviewer’s concern about X. We agree that Y must be addressed. We have not adopted the suggested method Z because [scientific reason]. Instead, we [alternative solution]. We have clarified this rationale in Methods/Discussion at lines…

56. Clarification template | Reviewer misunderstood because paper was unclear

We thank the reviewer for identifying this ambiguity. Our intention was X, but the original wording could reasonably be read as Y. We have revised the paragraph to define X explicitly and removed the ambiguous phrase.

57. Factual-correction template | Reviewer statement is incorrect

We appreciate the comment. We believe there may be a factual misunderstanding regarding X. The dataset includes Y rather than Z, as specified in Table 2. To make this easier to verify, we have added the relevant definition to Methods and revised the table caption.

58. Cannot-do template | Study cannot support requested experiment

We agree that this experiment would directly test the proposed mechanism. However, the necessary biospecimens/data were not collected and cannot be reconstructed retrospectively. We have therefore removed the mechanistic claim, reframed the explanation as a hypothesis and identify the experiment as a priority for future work.

59. New-analysis template | Added during revision

In response to the reviewer, we conducted an exploratory interaction analysis of baseline proficiency. This analysis was not prespecified, so we now label it exploratory in Methods and Results and avoid using it to alter the primary conclusion.

60. Part I operating rule | Revision begins with classification and dependency control

Do not ask only, “How do we answer this reviewer?” Ask, “What scientific object is this reviewer asking us to change, and what other parts of the manuscript depend on that object?”

不要只问：“这一条 reviewer comment 怎么回？”要问：“reviewer 要我改变的 scientific object 是什么？整篇 paper 哪些部分依赖这个 object？”

---

Part I establishes revision as controlled scientific change. Part II will classify difficult reviewer requests in depth—new analyses, new experiments, scope challenges, statistical disagreements, conflicting reviewers and comments that would make the paper worse if followed literally.

Part II — Difficult reviewer requests: what to change, what not to change, and how to preserve research history | 第二部分：最难处理的 reviewer requests

61. Reviewer asks for a new experiment that would answer a different research question

This is one of the most important revision decisions. A requested experiment may be scientifically interesting but conceptually external to the submitted paper.

Ask:

• Does it test the current paper’s central claim?
• Does it resolve a genuine validity concern?
• Does it introduce a new population, mechanism or technology?
• Would the new experiment require a new Introduction and new theory?
• Would it double the paper’s scope?

62. The “same paper or new paper?” test

If the requested work changes the core identity card—question, population, intervention, outcome, mechanism or domain—you may be looking at a new paper rather than a revision.

63. Example: reviewer requests a child sample for an adult study

Original paper:

advanced adult bilingual learners.

Reviewer request:

Please repeat the experiment with secondary-school students.

This would test transportability to a new population. It may be valuable, but it is not necessarily required to establish the adult-study claim.

A defensible response:

We agree that adolescent replication is important for generalisation. However, the present manuscript addresses advanced adult learners and we do not have an adolescent sample available within the revision period. We have therefore narrowed the generalisation language throughout the manuscript and now identify adolescent replication as a separate next study.

64. Example: reviewer requests a mechanism experiment

Original paper:

effect of structured feedback on revision performance.

Reviewer asks:

Manipulate cognitive load to prove the mechanism.

If mechanism was not the paper’s core claim, the scientifically correct revision may be to remove mechanistic certainty rather than launch a new experiment.

65. Reviewer requests an experiment that would rescue an overclaim | Sometimes reviewer reveals your original claim was too big

If the only way to defend your Discussion is to run a completely new mechanistic programme, the simpler repair may be to narrow the claim.

66. The “claim reduction before data expansion” principle

Before collecting new data, ask whether the reviewer’s concern can be resolved by making the paper claim only what the existing evidence supports.

67. New experiment required for validity vs new experiment required for ambition

Validity: without the control, the primary effect itself is uninterpretable.

Ambition: the primary effect is interpretable, but a broader mechanism/generalisation claim is not.

These should be treated differently.

68. Additional analysis request: first define the estimand/question

Do not run analyses because they sound statistically sophisticated. Ask what question the requested analysis answers.

69. Reviewer says “control for more variables”

Potential issues:

• new covariates may reduce confounding;
• new covariates may introduce overadjustment;
• colliders may create bias;
• sample size may not support more parameters;
• post-hoc model searching may increase researcher degrees of freedom.

70. Response to “add all available covariates”

A careful response can explain why covariates were chosen by design/causal rationale rather than significance testing or reviewer preference.

71. Reviewer asks for subgroup analyses | High risk of narrative drift

Before running:

• Was subgroup prespecified?
• Is interaction test possible?
• Is sample size adequate?
• How many subgroup analyses already exist?
• Will it be labelled exploratory?

72. Never convert reviewer-requested subgroup into original hypothesis

Introduction should not be rewritten to say:

We hypothesised a stronger effect among beginners.

if the hypothesis emerged during revision.

73. Reviewer requests mediation analysis | Mechanism trap

Ask whether temporal order and assumptions support causal mediation. If not, perform only if useful as exploratory association and label accordingly.

74. Reviewer requests machine-learning model comparison | Define the benchmark and split

New model comparisons can require re-running pipelines, changing preprocessing and updating figures. Treat them as scientific changes, not table additions.

75. Reviewer requests alternative statistical test | First identify why

Possible underlying concerns:

• distributional assumption;
• clustering;
• repeated measures;
• small sample;
• multiple testing;
• non-independence;
• model fit.

76. Do not answer “we followed standard practice” if the reviewer identifies a real assumption problem

Check the science.

77. Sensitivity analysis as compromise | Sometimes best response

Keep a prespecified primary model and add a reasonable alternative model to show whether the conclusion depends on analytic choice.

78. Sensitivity result changes direction | Revision becomes major

If a reasonable alternative analysis reverses the effect, you cannot continue to call the result robust.

Update:

• Results;
• Abstract;
• Discussion;
• Conclusion;
• possibly title.

79. Sensitivity result changes magnitude but not direction | Report the instability honestly

The direction was consistent, but estimated magnitude varied across specifications.

80. Reviewer requests exclusion of “outliers” | Ask whether rule is principled

Never remove observations simply because they weaken the result.

If an exclusion rule is defensible, report both the rationale and sensitivity where appropriate.

81. Reviewer requests removal of inconvenient cases | Protect analysis integrity

You can respectfully refuse an exclusion that has no methodological basis.

82. Reviewer requests transformation to achieve significance | Red flag

Analysis choices should be justified scientifically, not by whether p crosses a threshold.

83. Reviewer asks for post-hoc power calculation | Understand the purpose

Observed post-hoc power often adds little beyond the estimate and uncertainty. Depending on field conventions, explain why confidence intervals or precision are more informative.

84. Reviewer asks for “no difference” conclusion from non-significance

Do not comply if equivalence was not tested.

Respond with estimate/interval logic.

85. Reviewer asks you to claim causality from observational data

Do not increase causal strength merely to satisfy a reviewer. Explain design limits.

86. Reviewer asks you to remove causal language from a randomised trial | Check exact claim

You may defend a causal effect of assignment on the measured outcome while still narrowing mechanism or generalisation claims.

87. Reviewer says “the study is underpowered” | Clarify what is imprecise

Discuss estimate precision, sample-size planning and the specific claim affected. Do not answer with significance alone.

88. Reviewer says “sample too small” without specifying consequence

Ask or infer the likely concern:

• precision?
• subgroups?
• model complexity?
• generalisation?
• rare harms?

Then address the actual issue.

89. Reviewer requests more citations | Citation request taxonomy

Possible purposes:

• missing foundational literature;
• missing contradictory evidence;
• missing regional/disciplinary context;
• self-citation request;
• citation expansion without clear value.

90. Add citations when they improve fairness and context

Do not add irrelevant references simply to appease.

91. Reviewer requests their own work | Judge relevance, not identity

If relevant, cite it. If not, do not add it merely because the reviewer might be the author.

92. Reviewer challenges novelty | Replace vague novelty with specific contribution

Instead of defending:

This is novel.

state:

Earlier studies measure supported performance; this study adds an unseen delayed transfer task.

93. Reviewer says “not sufficiently novel” | Scientific response may be repositioning

Ask:

• Is contribution incremental but useful?
• Is novelty in method, population, validation or boundary?
• Was original Introduction overselling novelty?

94. Reviewer challenges generalisation | Often a scope correction

Do not answer with more adjectives. Restore exact sample and context.

95. Reviewer says “one site only” | Possible repairs

• narrow title;
• narrow abstract conclusion;
• add transportability limitation;
• avoid system-wide recommendation;
• propose multi-site replication.

96. Reviewer challenges outcome validity | Serious construct issue

If outcome does not measure the promised construct, the problem may require upstream repair in Introduction and Title.

97. Example: “engagement” measured by attendance only

Possible revision:

rename the measured outcome as behavioural attendance/participation and narrow claims about engagement.

98. Reviewer challenges terminology | Determine whether language or construct is wrong

Sometimes the term is merely unclear.

Sometimes the term changes the scientific claim.

99. Reviewer challenges Methods reproducibility | Add decision-relevant detail

Do not dump a lab notebook into Methods. Add the information necessary to understand and reproduce the study logic.

100. Reviewer requests raw data | Check ethics, consent, privacy and journal policy

Transparency is important, but data sharing can be constrained legitimately. Explain access conditions and provide de-identified/controlled-access options where appropriate.

101. Reviewer asks for code | If code is part of reproducibility, consider sharing/versioning

Ensure code corresponds to final manuscript analysis.

102. Reviewer identifies discrepancy between code and paper | Highest priority

Re-run analyses and propagate corrections. Do not treat as wording issue.

103. Reviewer requests a different theory | Do not rewrite the paper into reviewer’s preferred theoretical school automatically

Consider whether the theory explains the evidence better. If not, explain why your framing remains appropriate.

104. Reviewer says “add Theory B” | Possible good response

Add Theory B as a rival or boundary if relevant, without pretending it was the original framework.

105. Reviewer says your interpretation is wrong | Compare models

List:

• your interpretation;
• reviewer’s interpretation;
• evidence each explains;
• evidence that distinguishes them;
• what remains unresolved.

106. If both explanations remain plausible | Say so

Revision can improve the paper by moving from one-story certainty to model comparison.

107. Reviewer requests stronger limitations | Propagate them

A stronger limitation paragraph may require weaker Abstract/Conclusion claims.

108. Reviewer requests shorter paper | Cut repetition, not integrity

Protect:

• primary outcome definition;
• main estimate;
• critical limitation;
• planned/exploratory distinctions;
• response to editor concerns.

109. Reviewer requests more detail and editor requests shorter paper | Resolve by information architecture

Move technical detail to Supplement where appropriate while keeping core decisions visible in main text.

110. Reviewer requests restructuring | Ask what comprehension problem they experienced

Reordering can improve logic without changing science.

111. Reviewer says “Introduction too long” | Protect gap logic

Cut background that does not contribute to:

known → unknown → why it matters → question.

112. Reviewer says “Discussion too long” | Cut repetition of Results first

Protect:

• primary interpretation;
• strongest rival;
• critical limitation;
• scope;
• next discriminating study.

113. Reviewer says “Methods too long” | Move stable protocols, keep decisions

Do not cut the details that determine interpretation.

114. Reviewer says “Abstract too technical” | Simplify wording, not design identity

Do not simplify “randomised controlled trial” into “study” if that distinction matters.

115. Reviewer asks to remove a null result | Usually resist selective reporting

If the result is prespecified or materially affects interpretation, keep it.

116. Reviewer asks to move null result to supplement | Ask whether hierarchy remains fair

A primary or critical null result belongs in main text.

117. Reviewer asks to emphasise a positive secondary result | Preserve primary outcome priority

118. Reviewer asks to remove “exploratory” label | Do not erase history

If it was post hoc, keep the temporal status.

119. Reviewer requests new outcome after seeing results | Outcome switching risk

You can analyse it if scientifically useful, but label it as added during revision.

120. Reviewer requests redefinition of primary outcome | High-level editor discussion may be necessary

If the request materially changes the registered/declared primary outcome, be transparent and consider seeking editorial clarification.

121. Reviewer requests data exclusion that changes significance | Report sensitivity

Do not hide dependence on exclusion choices.

122. Reviewer requests a different missing-data method | Compare assumptions

A useful revision may present both primary and sensitivity analyses.

123. Reviewer requests correction for multiple testing | If multiplicity is real, address it

May require relabelling previously “significant” secondary findings.

124. Multiplicity correction can propagate to Abstract and Conclusion

If secondary results lose evidential strength, narrative emphasis must change.

125. Reviewer requests Bayesian analysis | Ask if it answers a useful question and can be implemented defensibly

Do not append a poorly specified Bayesian model for decoration.

126. Reviewer requests frequentist analysis of Bayesian paper | Same principle

Sensitivity can be helpful, but epistemic frameworks should not be mixed carelessly.

127. Reviewer requests nonparametric test | Check whether concern is assumptions or scale

128. Reviewer requests removal of covariate adjustment | Compare design rationale

129. Reviewer requests new control group | Determine whether primary interpretation otherwise fails

130. Reviewer requests measurement validation | If measure is central, this may be necessary

131. Reviewer requests language polishing | Separate style from scientific revisions

Do not let copyediting obscure major changes or introduce new construct synonyms.

132. Reviewer requests native-speaker editing | Focus on clarity, not prestige assumptions

Scientific English should be clear and accurate. Native-speaker status is not a quality criterion.

133. Reviewer requests “more confidence” in tone | Evidence controls confidence

134. Reviewer requests “less hedging” | Remove vague stacked hedges, not necessary uncertainty

135. Reviewer requests “more cautious” language | Check proposition size, not only hedge words

136. Reviewer asks for more figures | Each figure needs a question

137. Reviewer asks to combine figures | Ensure sample/time/metric differences remain visible

138. Reviewer asks to move figure to supplement | Keep primary evidence visible

139. Reviewer requests graphical abstract | Audit visual causal arrows and scope

140. Reviewer requests lay summary | Plain language must preserve evidence strength

141. Reviewer requests title change | Re-run title–abstract–conclusion coherence

142. Editor requests title change | Still protect scientific scope

143. Reviewer requests new keyword | SEO/metadata should not broaden scientific claim

144. Reviewer requests more policy implications | Add only if decision evidence exists

145. Reviewer asks for “clinical/practical significance” | Use a benchmark or decision criterion

146. Reviewer requests effect size | Often improves Results interpretability

147. Reviewer requests confidence intervals | Often improves precision reporting

148. Reviewer requests p-values only | Do not remove effect estimates if they are important

149. Reviewer asks to “prove” hypothesis | Avoid proof language in empirical contexts unless warranted

150. Reviewer asks for stronger conclusion after extra analysis | Reassess whole evidence state

New evidence can justify stronger wording, but only after:

• analysis quality check;
• multiplicity considerations;
• scope review;
• updated limitation analysis.

151. Revision triage matrix | Importance × validity × effort

Concern importance	Requested fix valid?	Response
high	yes	prioritise and propagate fully
high	no	solve underlying concern another way
low	yes	make if useful and low-cost
low	no	politely decline or clarify

152. Effort alone should not decide | Science first

A difficult but validity-critical revision deserves priority. An easy but invalid change should not be made.

153. Deadline pressure | Ask editor early if valid revision needs more time

Do not rush a critical experiment or analysis into an unreliable form solely to meet a deadline. Journal-specific policies vary; communicate with the editor where necessary.

154. Revision feasibility meeting | Before work begins

Team should decide:

• what must change;
• what can change;
• what cannot be done;
• what new uncertainty arises;
• what deadlines/resources exist;
• who owns each dependency.

155. The “reviewer requested it” fallacy | External request does not remove author responsibility

Every added analysis, exclusion and claim remains yours.

156. The “we did everything” fallacy | Compliance can make a worse paper

A revision that blindly adds all requested analyses can become statistically incoherent and narratively fragmented.

157. The “defend everything” fallacy | Not every original choice deserves defence

Some reviewer comments reveal real weaknesses. Change them.

158. The “reviewer is wrong, therefore no manuscript change” fallacy

Even if the reviewer is factually wrong, the manuscript may have enabled the misunderstanding.

159. The “major revision means rejection” fallacy

A major revision is an invitation to improve and resubmit, not an acceptance guarantee and not automatically a rejection.

160. Part II operating rule | Protect scientific validity before author pride or reviewer preference

The best revision is not the one that obeys the most comments. It is the one that resolves the strongest concerns while leaving the final manuscript scientifically more coherent than the submitted version.

最好的 revision，不是“改得最多”，而是解决最重要的问题以后，整篇 paper 比原稿更 coherent、更透明、更能守住 evidence。

---

Part II classified the hard requests. Part III will now show how one accepted revision propagates through the whole manuscript—Title, Abstract, Introduction, Methods, Results, figures, Discussion and Conclusion—without leaving contradictory versions behind.

Part III — Propagate every accepted change through the whole manuscript | 第三部分：一个修改怎样传遍整篇论文

161. Revision dependency graph | One scientific object, many manuscript surfaces

Every important manuscript object has dependents. Change the object, and its dependents must be audited.

Examples:

Primary outcome → Abstract → Methods → Results order → Discussion emphasis → Conclusion.

Population definition → Title → Abstract → Methods → generalisation language → Conclusion.

Analysis choice → Methods → Results estimates → figures → uncertainty → Discussion → Conclusion.

162. Dependency graph for a revised research question

If a reviewer shows that the original research question is too broad, revise:

• Title if it encodes the broad question;
• Abstract objective;
• Introduction gap;
• Methods statement of objective;
• Results hierarchy;
• Discussion opening;
• Conclusion.

163. Example: “durable learning” becomes “four-week persistence”

Reviewer correctly notes that four weeks is insufficient to support broad long-term language.

Change propagation:

1. Title: remove “lasting”.
2. Abstract: replace “durable learning” with “four-week persistence”.
3. Introduction: narrow gap from long-term learning to short delayed transfer.
4. Results: no change to numbers.
5. Discussion: distinguish measured persistence from longer-term durability.
6. Conclusion: state four-week evidence and longer-term uncertainty.

164. Scientific revision can leave Methods unchanged while changing interpretation | Not every dependency touches every section

The goal is not to edit everything automatically. It is to inspect everything that depends on the changed object.

165. Dependency graph for a corrected construct label

Reviewer:

“Your scale measures self-efficacy, not competence.”

Audit:

• Title;
• Abstract;
• Introduction theory;
• Methods scale definition;
• Results variable labels;
• Figures/axes;
• Discussion interpretation;
• Conclusion.

166. Do not use global find–replace without conceptual review

“Competence” and “self-efficacy” are not synonyms. Replacing strings does not guarantee the surrounding argument still works.

167. Dependency graph for sample-size correction

Reviewer finds that Abstract says n = 126 but 120 were actually randomised.

Audit:

• Abstract;
• Methods recruitment flow;
• CONSORT/flow diagram;
• Results denominators;
• table footnotes;
• percentages;
• Supplement;
• possibly power/sample-size explanation.

168. One corrected n can change percentages | Recalculate, do not just replace

If numerator remains 90:

90/126 ≠ 90/120.

169. Dependency graph for an altered exclusion rule

If revision changes who is analysed, update:

• Methods exclusion rule;
• participant flow;
• analysis n;
• descriptive tables;
• primary estimates;
• figures;
• Discussion limitation;
• response letter.

170. Re-running is mandatory when the analytical population changes

Do not edit counts manually while leaving estimates from the old sample.

171. Dependency graph for a new missing-data method

Reviewer asks for multiple imputation as a sensitivity analysis.

Add/update:

• Methods missing-data subsection;
• Results sensitivity estimate;
• possibly a table;
• Discussion robustness language;
• Conclusion only if interpretation changes.

172. If sensitivity analysis confirms primary result | Confidence may increase modestly

Write:

The estimate was similar under multiple imputation, reducing concern that the primary direction depended on complete-case analysis.

173. If sensitivity analysis weakens primary result | Narrative must change

Do not hide the sensitivity analysis in Supplement while retaining “robust” in main text.

174. Dependency graph for multiplicity correction

If reviewer correctly notes that many secondary tests were uncorrected:

• Methods statistical plan;
• Results p-values/interval interpretation;
• table significance markings;
• Discussion secondary claims;
• Abstract if secondary outcome highlighted;
• Conclusion if affected.

175. A significance star is not an isolated formatting object | It can drive narrative

Removing a star may require removing “significant improvement” from prose.

176. Dependency graph for a new covariate-adjusted model

Reviewer raises baseline imbalance.

If adjusted model becomes primary:

• Methods define model;
• Results primary estimate changes;
• Abstract result changes;
• figure/table labels change;
• Discussion should interpret adjusted result;
• unadjusted result may become descriptive or sensitivity.

177. Do not mix old unadjusted and new adjusted estimates across sections

This creates a paper with two incompatible “primary results.”

178. Dependency graph for a new subgroup analysis

Reviewer-requested subgroup:

• Methods: label exploratory/post hoc;
• Results: estimate interaction and uncertainty;
• Discussion: hypothesis-generating interpretation;
• Conclusion: usually no upgrade to primary claim;
• Introduction: do not rewrite as original hypothesis.

179. New subgroup finding can generate a future hypothesis | That is a legitimate contribution

It simply remains temporally honest.

180. Dependency graph for a new mechanism measure

Suppose revision adds a process measure already collected but omitted.

Audit:

• Methods measure;
• analysis;
• Results;
• Discussion mechanism interpretation;
• Abstract only if central;
• Conclusion only if mechanism evidence materially strengthens.

181. Adding an existing but previously unreported measure raises reporting questions

Explain why it was added and whether it was prespecified.

182. Dependency graph for a new experiment | Treat as a new evidence node

A new experiment can require:

• Introduction scope expansion;
• new Methods subsection;
• new Results subsection;
• new figures;
• restructured Discussion;
• updated Abstract;
• possibly new Title.

183. New experiment should not erase original experiment hierarchy

If Experiment 2 is added during revision, clarify whether it confirms, extends or tests a mechanism from Experiment 1.

184. Revision-added experiment can become central | But chronology still exists

Response letter can state:

In response to Reviewer 1, we added Experiment 2 to test the mechanism directly.

185. Manuscript Introduction can integrate Experiment 2 without pretending it existed before review

The published paper need not narrate the peer-review timeline unless relevant, but the response letter should.

186. Dependency graph for a narrowed causal claim

Reviewer correctly notes observational design.

Search globally for:

• causes;
• leads to;
• improves;
• results in;
• prevents;
• because of X.

Then inspect each occurrence rather than replacing blindly.

187. Causal drift often survives in figure titles and captions

Check visuals, graphical abstract and highlights.

188. Dependency graph for a strengthened causal claim

If reviewer overlooked randomisation and asks you to weaken the effect claim, you may defend the causal effect of assignment while clarifying what remains non-causal:

• mechanism;
• generalisation;
• component attribution.

189. Causality is multidimensional | Treatment effect can be causal; mechanism can remain uncertain

190. Dependency graph for a new limitation

Reviewer identifies attrition bias.

Update:

• Results participant flow;
• possibly sensitivity analysis;
• Discussion limitation;
• scope/precision;
• Abstract or Conclusion if materially affected.

191. Limitations should propagate only as far as their consequence

Do not weaken unrelated claims.

192. Dependency graph for a literature correction

Reviewer identifies a contradictory high-quality study omitted from Introduction.

Update:

• Introduction evidence state;
• gap framing;
• Discussion comparison;
• possibly contribution claim.

193. New literature can shrink novelty | Update novelty honestly

If another study already did something similar, do not retain “first” language.

194. Dependency graph for a contribution change

Original contribution:

first delayed test.

Reviewer supplies earlier delayed study.

New contribution:

first delayed test in advanced bilingual adults with an unseen task.

Audit title/abstract/introduction/discussion/conclusion.

195. Dependency graph for figure correction

Reviewer detects axis error.

Check:

• underlying plotted data;
• caption;
• Results prose;
• Discussion pattern;
• Abstract/Conclusion if affected.

196. Never assume figure correction is visual only

197. Dependency graph for table correction

Incorrect denominator or coefficient can alter downstream interpretation.

198. Dependency graph for revised terminology

Reviewer asks to distinguish “efficacy” and “effectiveness.”

Check every use because the terms encode different settings and claims.

199. Dependency graph for revised population label

“Asian students” is too broad; actual sample is Singapore-based advanced bilingual adults.

Update all generalisation surfaces.

200. Dependency graph for revised time language

“Long-term” → “12-week.”

Search:

• Title;
• Abstract;
• Introduction;
• figure titles;
• Discussion;
• Conclusion.

201. Dependency graph for revised outcome hierarchy

If editor requires primary outcome to be clearly identified:

• Introduction objective;
• Methods;
• Results order;
• Abstract order;
• Discussion opening;
• Conclusion.

202. Dependency graph for revised sample-flow reporting

If attrition clarified:

• flow diagram;
• Methods;
• Results;
• table n;
• limitation;
• response.

203. Dependency graph for revised reporting guideline compliance

A CONSORT/STROBE/PRISMA checklist may reveal missing items. Add them where they belong rather than creating one generic “reporting” paragraph.

204. Change log | One row per scientific change

Change ID	Trigger	Scientific object	Sections affected	Completed?
C1	R1.3	missing-data analysis	M,R,D,C,A	yes
C2	R2.1	construct label	T,A,I,M,R,D,C	yes
C3	E2	generalisation scope	T,A,D,C	yes

Key: T = Title; A = Abstract; I = Introduction; M = Methods; R = Results; D = Discussion; C = Conclusion.

205. Response matrix and change log are different | Comment vs scientific change

One reviewer comment may create several scientific changes.

Several reviewer comments may be solved by one scientific change.

206. Example: three reviewers question overclaiming | One global scope correction

Instead of making three local wording patches, narrow the central claim once and propagate consistently.

207. Version-control naming | Avoid final_FINAL_v7 chaos

Use a simple convention:

Manuscript_R1_clean_2026-09-18.docx

Manuscript_R1_tracked_2026-09-18.docx

Response_R1_2026-09-18.docx

208. Lock a revision baseline | Save submitted version unchanged

Never overwrite the exact version that reviewers saw.

209. Keep analysis outputs versioned | Manuscript numbers must map to a reproducible analysis state

210. Re-run final analyses from clean pipeline where possible

This catches partial manual changes.

211. Manuscript change order | Science before style

1. Data/analysis corrections.
2. Methods/Results updates.
3. Figures/tables.
4. Discussion/Conclusion.
5. Introduction/Abstract/Title.
6. Language/style.
7. Response letter finalisation.

212. Why Abstract and Title are late in the sequence

They should inherit the final revised evidence state.

213. Draft response while making changes | Do not reconstruct everything at end

As each comment is resolved, record:

• decision;
• analysis/change;
• manuscript location;
• revised text;
• dependencies checked.

214. But finalise response only after whole-paper audit

Otherwise the response can describe an intermediate version that changed again later.

215. Response line numbers should match final tracked manuscript

Re-number after all changes are complete.

216. Cross-comment interaction | One comment can make another obsolete

Reviewer 1 asks to explain a mechanism.

Reviewer 2 correctly shows mechanism cannot be identified.

If authors remove mechanism claim, Reviewer 1’s requested explanatory paragraph may no longer be needed.

217. Resolve comments as a system, not in chronological order

Prioritise by scientific dependency:

1. validity;
2. analysis;
3. construct/scope;
4. interpretation;
5. clarity;
6. style.

218. Editor comments can create umbrella changes

If editor says “the claim is too broad,” multiple reviewer comments may collapse into one scope revision.

219. Re-review risk | Reviewer sees old comment and new paper side by side

Do not claim a change was made if manuscript does not visibly contain it.

220. Response–manuscript mismatch is a serious credibility problem

Example:

Response:

We now label the subgroup exploratory.

Manuscript:

As hypothesised, lower-baseline learners benefited more.

This is contradictory.

221. The response letter should be audited against the manuscript | Every “we changed” needs verification

222. The manuscript should be audited against the response | Every substantive revision should be documented

223. Tracked-changes overload | Major rewrites can become unreadable

Follow journal instructions. When a document becomes difficult to assess, provide a clean version plus a well-organised response and marked version as requested.

224. Deleted text matters too | Response should explain important removals

If you remove a mechanism claim, state that clearly.

225. New limitations can alter the “selling point” of the paper | That is acceptable

Revision should prioritise truth over preservation of marketing language.

226. New analyses can strengthen the paper | Avoid becoming triumphalist

Even if robustness improves, retain original design boundaries.

227. New analyses can weaken the paper | That must propagate too

Peer review sometimes reveals that the submitted claim was too strong. A successful revision may be a narrower paper.

228. Narrower can be publishable because it is more reliable

229. Reviewer-requested data can fail to replicate the original effect

Do not omit the new evidence. The revision may need a fundamentally changed Discussion.

230. New experiment contradicts original result | Revision architecture

Possible response:

• report both experiments;
• analyse design/population differences;
• reduce certainty;
• reframe contribution around boundary conditions or inconsistency.

231. A contradictory replication can become the paper’s strongest contribution

Science does not require the revision to preserve the original story.

232. The “story preservation” bias | Authors often unconsciously protect submission narrative

Revision should protect evidence integrity, not the first draft’s storyline.

233. Add a “what changed scientifically?” memo for the team

One page:

• old primary claim;
• new primary claim;
• new analyses;
• new data;
• new limitations;
• changed contribution;
• unchanged core findings.

234. This memo prevents co-authors from editing toward the old paper

235. Revision-round identity card | Re-freeze after major changes

Create:

Study Identity R1

then, if needed:

Study Identity R2.

236. Track differences between identity cards

If R1 changed the primary claim, R2 authors should know.

237. Multi-round review can accumulate contradictions | Round 2 reviewer may see manuscript after Round 1 changes

Do not treat R2 as if submission version still exists.

238. Reviewer requests reversal of previous reviewer-requested change | Explain history to editor if needed

Do not oscillate silently.

239. Example: R1 asks to add covariate; R2 asks to remove it

Response should explain scientific rationale and possibly present sensitivity analyses rather than changing primary model repeatedly.

240. Part III operating rule | Every scientific change has dependents

A revision is complete only when the changed scientific object, all manuscript surfaces that depend on it, and the response document all agree on the same new state.

一个 revision 真正完成，不是某一句改好了，而是 scientific object 改完以后，所有依赖它的 section、figure、table、Abstract、Conclusion 与 response letter 都进入同一个新状态。

---

Part III built the propagation system. Part IV will build the response-letter system itself: point-by-point architecture, disagreement language, evidence-based rebuttal, editor-only cover-letter issues, line references, and examples of strong versus weak replies.

Part IV — Write a response letter reviewers can actually assess | 第四部分：把 rebuttal 写成 reviewer 能快速验证的文件

241. The response letter is not a second manuscript | It is an audit trail

The best response document lets an editor or reviewer answer three questions quickly:

1. Did the authors understand the concern?
2. What did they change or decide not to change?
3. Where can I verify that change?

Everything else is secondary.

242. Response architecture | Four layers

1. Brief opening.
2. Summary of major revisions.
3. Point-by-point responses.
4. Closing note.

243. Opening paragraph | Short, professional, specific

Useful pattern:

We thank the editor and reviewers for their careful assessment and the opportunity to revise the manuscript. We have addressed all comments and made several substantive changes, including [A], [B] and [C]. Below we provide a point-by-point response and identify the location of each revision in the manuscript.

244. Do not spend half a page praising the reviewers | Respect does not require ceremony

A brief thanks is enough. The response document’s value comes from clarity and completeness.

245. Major-revision summary | Name scientific changes, not editing volume

Strong summary:

• re-analysed primary outcome with clustered standard errors;
• added multiple-imputation sensitivity analysis;
• narrowed causal claims to match observational design;
• revised outcome terminology from competence to self-efficacy;
• added a single-site transportability limitation;
• updated Abstract and Conclusion accordingly.

246. Weak revision summary

We made extensive improvements throughout the manuscript.

This tells the editor almost nothing.

247. Point-by-point response | Repeat the full comment where feasible

Nature Computational Science recommends preserving each reviewer comment in full before the response. This reduces selective response and makes context visible.

Nature Computational Science | How to Respond to Reviewers

248. Distinguish reviewer text from author text visually

Use formatting appropriate to the journal:

• bold reviewer comment;
• plain author response;
• indented quotation;
• different font style;
• clear labels.

249. One comment, one response block | Avoid combining unrelated requests

If one reviewer paragraph contains three distinct requests, split internally:

R1.4a, R1.4b, R1.4c.

250. Response block template | Default structure

Reviewer comment

Full comment.

Response

What you think the concern is and whether you agree.

Change made

What changed scientifically/textually.

Location

Page/line/section/table/figure.

251. Example: straightforward agreement

Reviewer: “Please define how the primary outcome was scored.”

Response: We agree that the original Methods did not provide enough scoring detail. We have added the rubric dimensions, score range, rater training procedure and disagreement rule.

Location: Methods, “Outcome Measurement,” lines 221–238.

252. Example: agreement that changes interpretation

Reviewer: “The term ‘durable learning’ is too strong for a four-week follow-up.”

Response: We agree. We have replaced “durable learning” with “four-week persistence” throughout the manuscript and now state explicitly that longer-term durability was not assessed. This change affects the Title, Abstract, Discussion and Conclusion.

253. Example: partial agreement

Reviewer: “Because groups differ at baseline, please use propensity-score matching.”

Response: We agree that baseline imbalance requires explicit adjustment. We have not used propensity-score matching because treatment assignment was randomised and the imbalance is consistent with chance variation. Instead, we now present the prespecified baseline-adjusted regression as the primary estimate and add the unadjusted comparison as descriptive context. We also clarify this rationale in Methods.

254. Partial agreement shows you heard the concern | You do not have to copy the proposed fix

255. Example: respectful disagreement

Reviewer: “Please remove the null primary outcome because it is not statistically significant.”

Response: We respectfully disagree with removing the primary outcome. It was prespecified and directly addresses the study’s main question. Omitting it would distort the evidence hierarchy. We have instead revised the Results to report the estimate and confidence interval more clearly and expanded the Discussion of its uncertainty.

256. Strong disagreement uses scientific principles, not authority

Weak:

We disagree because this is standard in our laboratory.

Stronger:

We disagree because the requested exclusion rule is not prespecified, lacks a measurement-based rationale and materially changes the estimate. We therefore retain all eligible observations and provide the requested exclusion only as a sensitivity analysis.

257. Example: reviewer factual misunderstanding + manuscript clarification

Reviewer: “The control group received no feedback.”

Response: The control group did receive feedback; it received standard evaluative feedback without action prompts. We recognise that the original wording did not make this comparator sufficiently clear. We have revised the Abstract and Methods to define the control condition explicitly and updated Figure 1 accordingly.

258. Notice the two-part response | Correct reviewer + improve manuscript

259. Example: cannot perform new experiment

Reviewer: “Please repeat the study with a six-month follow-up.”

Response: We agree that six-month follow-up would provide a stronger test of durability. However, the study has ended and additional follow-up data cannot be collected. We have therefore narrowed all long-term claims, removed “durable” from the Title and Conclusion, and identify six-month replication as the next critical study.

260. Cannot-do responses should show scientific repair | Not just inability

If you cannot meet the requested experiment, remove or narrow the claim that required it.

261. Example: new post-hoc analysis

Reviewer: “Could the effect differ by baseline proficiency?”

Response: We conducted an exploratory group-by-baseline interaction analysis in response to this comment. The interaction estimate suggests a larger effect among lower-baseline learners but is imprecise. Because this analysis was not prespecified, we label it exploratory in Methods, Results and Discussion and do not use it to alter the primary conclusion.

262. Example: new sensitivity analysis

Reviewer: “Please examine whether missing-data assumptions affect the result.”

Response: We agree. We added a multiple-imputation sensitivity analysis under a missing-at-random assumption. The estimate remained similar to the complete-case result. We now describe the method in Methods, report the estimate in Results and revise the Discussion to state that the primary direction was not materially changed by this alternative missing-data approach.

263. Example: new analysis weakens result

Reviewer: “Please account for clustering by instructor.”

Response: We agree and have re-estimated the primary model with instructor-level clustered standard errors. The point estimate is unchanged, but the confidence interval is wider and now includes smaller effects. We have updated the Abstract, Results, Figure 2, Discussion and Conclusion to reflect the increased uncertainty.

264. This is what scientific revision looks like | The response does not hide the weaker evidence

265. Example: reviewer requests stronger claim but authors retain caution

Reviewer: “The randomised design supports stronger language; please say the intervention causes improved independent writing.”

Response: We agree that random assignment supports a causal effect of assignment to the structured condition on the measured one-week revision outcome. We have strengthened that statement accordingly. We retain more cautious language for “independent writing ability,” because the study measured revision on one unseen task rather than the broader construct of writing ability.

266. Good response separates what can be strengthened from what cannot

267. Example: reviewer asks for unnecessary certainty downgrade

Reviewer: “Please change ‘improved’ to ‘was associated with’ throughout.”

Response: We have retained causal language for the effect of random assignment on the prespecified outcome because the study used random allocation and the primary analysis estimates the between-condition effect. We have, however, revised the Discussion to make clear that the mechanism and generalisation beyond the studied sample remain uncertain.

268. Response language for conflicting reviewers

Reviewer 1 recommends removing the theoretical section, whereas Reviewer 2 asks for a fuller comparison with Theory B. We addressed both concerns by shortening the general theory review while adding a focused paragraph comparing the two mechanisms directly relevant to the findings.

269. Do not tell Reviewer 1 that Reviewer 2 “won” | Frame as editorial solution

270. Response language for editor override

Following the editor’s request to prioritise the primary outcome, we retained the main Results order and moved the reviewer-requested exploratory subgroup analysis to the end of the Results section and Supplementary Figure S3.

271. Response to unclear comment | Ask editor or interpret cautiously

If comment is too vague:

“The analysis is weak.”

Possible response:

We interpreted this comment as a concern about model specification and have now added the assumption checks and sensitivity analysis described below. If the reviewer intended a different concern, we would be grateful for clarification.

272. When to contact editor before revision | High-value cases

• reviewers demand mutually exclusive major experiments;
• request would violate ethics/consent;
• request requires inaccessible proprietary data;
• review contains serious factual or conflict concerns;
• requested scope is far beyond submitted work;
• deadline is incompatible with necessary valid work.

273. Editor-only cover letter | Different from reviewer response

Nature Computational Science notes that concerns about review reports can be raised in a cover letter to the editor, which is not shared with reviewers in that context.

Use editor-only communication for:

• confidential conflict concerns;
• inappropriate reviewer conduct;
• requests that exceed scope dramatically;
• ethical barriers;
• requests for editorial clarification.

274. Do not use reviewer-facing response to accuse reviewer of bias | Escalate appropriately

275. Tone ladder | Neutral, appreciative, evidence-based

Useful verbs:

• agree;
• clarify;
• revise;
• add;
• remove;
• reanalyse;
• retain;
• respectfully disagree;
• cannot assess;
• cannot exclude.

276. Avoid defensive verbs

Examples to avoid:

• obviously;
• clearly the reviewer misunderstood;
• we already said;
• as any expert knows;
• we refuse;
• the reviewer is wrong.

277. “We already stated this” is rarely enough | Make it easier to find

If a reviewer missed it, consider moving, repeating briefly or clarifying it.

278. “No change made” needs a reason | Every refusal should be inspectable

Strong:

No manuscript change was made because the requested analysis conditions on a post-treatment variable and would not estimate the same causal quantity as the primary analysis. We have added a sentence clarifying this rationale in Methods.

279. Sometimes disagreement still leads to manuscript improvement

You can reject the proposed analysis but add a clearer rationale so future readers understand the choice.

280. Quote revised text when wording itself resolves concern

Reviewer:

“The primary outcome is unclear.”

Response can include the revised sentence.

281. Do not quote revised text when concern is primarily analytical

For a new model, report the scientific result and location rather than pasting pages of code-like detail.

282. Page and line references | Use final numbering

Example:

Methods, p. 8, lines 214–229.

283. Section references can be more robust than page numbers in online systems

Follow journal preference.

284. Table/figure references | Be exact

New sensitivity analysis: Table S4.

285. Response length | Complete but not theatrical

A response can be longer than the manuscript if many comments require technical explanation, but each point should remain focused.

286. One-paragraph response test | Does first sentence state agreement/disagreement?

Reviewers should not have to read 500 words before learning your answer.

287. Evidence-first disagreement | Structure

1. acknowledge concern;
2. state position;
3. give scientific reason;
4. state manuscript change or why none;
5. give location.

288. Example of weak rebuttal | Too emotional

We strongly disagree with this reviewer, who appears not to understand our methodology. We have used this analysis for years and believe no change is needed.

289. Repair weak rebuttal

We appreciate the concern regarding model choice. The requested alternative estimates a different target quantity because it conditions on a post-treatment variable. We therefore retain the prespecified model and have added a sensitivity analysis that addresses the reviewer’s concern without changing the estimand. The rationale and sensitivity result are now reported in Methods and Results.

290. Example of weak agreement | Too empty

We agree and have changed the manuscript accordingly.

291. Repair empty agreement

We agree that the original Discussion implied broader generalisation than the single-site design supports. We now restrict the claim to the studied high-support programme and add multi-site replication as the next test. Corresponding changes were made in the Abstract, Discussion and Conclusion.

292. Response to grammar/style corrections | Batch where appropriate

For dozens of minor copyedits, you can acknowledge them collectively if journal/editor accepts this, while still addressing substantive comments individually.

293. Response to typo | No need for essay

Corrected. Thank you.

294. Response to unclear figure | Explain design change

We agree that the original figure obscured the group labels and error-bar definition. We have redesigned Figure 3, labelled each condition directly and defined 95% confidence intervals in the caption.

295. Response to over-dense table

Move secondary columns to Supplement but preserve primary result in main table.

296. Response to “please cite…” | Relevance test

If relevant:

We thank the reviewer for pointing us to this study. We have added it to the Introduction and Discussion because it provides a delayed-transfer comparison directly relevant to our contribution.

If not relevant:

We reviewed the suggested paper carefully. Its primary focus is X rather than Y, and we do not believe adding it would clarify the specific evidence gap addressed here. We have therefore not added the citation.

297. Never imply you know reviewer identity | Treat suggested citation neutrally

298. Response to reviewer praise | Do not over-focus

Acknowledge briefly and move to substantive issues.

299. Response to contradictory reviewer claims about literature

Return to sources. Your response should be evidence-based rather than choosing the friendlier reviewer.

300. Response to “not enough novelty” with added analysis | Do not manufacture novelty

Use the new analysis to sharpen contribution only if scientifically legitimate.

301. Response to “study too descriptive” | Clarify intended research job

If the study is deliberately descriptive, defend that job and its value rather than pretending it is causal.

302. Response to “make it causal” | Design controls claim

303. Response to “add policy recommendations” | Evidence controls decision scope

304. Response to “remove limitations to improve impact” | Do not trade credibility for apparent impact

305. Response to reviewer asks for longer Discussion but editor sets word limit | Use supplement or prioritise

306. Response document formatting | Consistency helps assessment

Use one stable pattern for all comments.

307. Number every reviewer comment | Especially long reviews

308. Preserve reviewer numbering if provided | Add subnumbers only where needed

309. Cross-reference related comments | Avoid duplicate explanations

See also our response to R1.3, where we describe the same sensitivity analysis.

310. But every comment still needs an explicit answer | Do not only say “see above”

311. Editor summary should mention conclusion-changing revisions | High importance

If the primary conclusion changed, say so clearly.

312. Do not claim “minor changes” when scientific conclusion changed

313. Cover letter vs response letter | Different audiences

Cover letter:

• editor-facing overview;
• confidential concerns;
• major revision summary;
• requested contextual information.

Response:

• reviewer/editor comment-by-comment audit trail.

314. Nature Communications and other Nature journals often request both cover letter and point-by-point response

Nature Communications | Editorial Process

315. Scientific Reports likewise requires point-by-point explanation for revised manuscripts

Scientific Reports | Peer Review and Revisions

316. Response letter final audit | Comment coverage

Every numbered comment has:

• response;
• decision;
• change/no-change rationale;
• location;
• status closed.

317. Response letter final audit | Manuscript consistency

Every claimed change actually exists.

318. Response letter final audit | Tone

No sarcasm, accusation, passive aggression or emotional language.

319. Response letter final audit | Scientific honesty

New analyses are labelled according to when they were added.

320. Part IV operating rule | Make it easy to verify that the paper improved

The response letter should reduce reviewer effort. A reviewer should be able to see the concern, your scientific decision, the resulting manuscript change and the exact place to verify it without reconstructing your revision process themselves.

好的 response letter，不是“写得客气”而已，而是让 reviewer 不需要自己侦查：comment 是什么、你怎么判断、你改了什么、在哪里能验证。

---

Part IV builds the response document. Part V will run the system on a complete fictional major revision, including conflicting reviewers, a new sensitivity analysis, a narrowed claim, one respectful refusal and a response letter that remains coherent with the revised manuscript.

Part V — Complete major-revision case: from decision letter to resubmission | 第五部分：完整 Major Revision 案例

All study details and review comments below are fictional teaching material.

321. Submitted manuscript identity | 原稿

Study:

120 advanced bilingual adult learners randomised to structured action-oriented feedback or standard evaluative feedback.

Primary outcome:

one-week independent revision score.

Secondary:

confidence, evidence-integration subscore, sentence accuracy, 12-week score.

Submitted primary result:

adjusted mean difference 5.7 points, 95% CI 2.3–9.1.

Submitted title:

Structured Feedback Builds Lasting Independent Writers

The title already overclaims durability and construct scope.

322. Decision letter | Major revision

Editor summary:

1. Primary claim appears broader than measured outcome.
2. Reviewer 1 raises clustering and missing-data concerns.
3. Reviewer 2 questions mechanism language and generalisation.
4. Both reviewers request clearer participant flow and stronger comparison with prior delayed-transfer literature.

323. Reviewer 1 comment R1.1 | Clustering

“Participants were taught by six instructors. The analysis appears to treat observations as independent. Please account for clustering by instructor.”

324. Reviewer 1 comment R1.2 | Missing data

“The one-week outcome is missing for eight participants and the 12-week outcome for more. Please assess sensitivity to missing-data assumptions.”

325. Reviewer 1 comment R1.3 | Subgroup

“The effect may be stronger for lower-proficiency learners. Please stratify by baseline proficiency.”

326. Reviewer 1 comment R1.4 | Conclusion

“The Conclusion could be stronger given randomisation.”

327. Reviewer 2 comment R2.1 | Mechanism

“The manuscript repeatedly claims that action clarity improves metacognition, but metacognition was not measured.”

328. Reviewer 2 comment R2.2 | Generalisation

“The sample comes from one high-support adult programme. Claims about bilingual learners broadly are not justified.”

329. Reviewer 2 comment R2.3 | Long-term language

“A 12-week follow-up does not establish durable writing ability. Please revise the framing.”

330. Reviewer 2 comment R2.4 | Literature

“The Discussion omits two delayed-transfer studies that report weaker persistence.”

331. Reviewer 2 comment R2.5 | New experiment request

“Please repeat the study in secondary-school students to demonstrate educational generalisability.”

332. Step 1 — Build the revision control sheet

ID	Concern	Decision	Scientific action	Dependencies
E1	claim too broad	accept	narrow central claim	T,A,I,D,C
R1.1	clustering	accept	clustered SE sensitivity/primary reassessment	M,R,F,A,D,C
R1.2	missing data	accept	multiple-imputation sensitivity	M,R,S,D
R1.3	subgroup	partly accept	exploratory interaction analysis	M,R,D,S
R1.4	stronger Conclusion	partly accept	strengthen treatment-effect wording, retain scope limits	D,C,A
R2.1	mechanism unsupported	accept	remove metacognition claim	A,D,C
R2.2	generalisation broad	accept	restrict population claim	T,A,D,C
R2.3	durability overclaim	accept	replace durable with 12-week persistence	T,A,I,D,C
R2.4	literature omission	accept	add delayed-transfer studies	I,D
R2.5	secondary-school replication	cannot address	narrow generalisation + future study	D,C,response

333. Step 2 — Re-run clustered analysis | Scientific result changes uncertainty

Original:

5.7 points, 95% CI 2.3–9.1.

Cluster-adjusted:

5.7 points, 95% CI 1.1–10.3.

The point estimate is unchanged; uncertainty is larger.

334. Propagation of R1.1

• Methods: describe instructor-cluster adjustment.
• Results: replace primary interval.
• Figure 2: update interval.
• Abstract: update interval if reported.
• Discussion: avoid calling magnitude precise.
• Conclusion: direction remains supported; magnitude less precise.

335. Step 3 — Multiple-imputation sensitivity

Imputed estimate:

5.3 points, 95% CI 0.7–9.9.

Direction remains similar.

336. Propagation of R1.2

Methods gets imputation assumptions.

Results gets sensitivity estimate.

Discussion states that direction did not materially depend on complete-case analysis, while uncertainty remained substantial.

Conclusion does not need a separate imputation sentence because the primary conclusion does not change.

337. Step 4 — Exploratory proficiency interaction

Interaction suggests larger effect at lower baseline proficiency but interval is wide.

338. Correct treatment of R1.3

Methods:

At Reviewer 1’s suggestion, we added a post-hoc exploratory group-by-baseline-proficiency interaction analysis.

Published manuscript can simply say:

An exploratory post-hoc interaction analysis examined whether the group difference varied by baseline proficiency.

Results:

report estimate and uncertainty.

Discussion:

hypothesis-generating only.

Introduction:

unchanged.

339. Step 5 — Stronger conclusion request | R1.4

Reviewer is partly right: randomisation supports a causal effect of assignment to the full structured condition on the measured one-week outcome.

But “develops independent writers” remains too broad.

Revision:

Assignment to structured feedback improved the measured one-week independent-revision outcome in this sample.

Mechanism, generalisation and durability remain limited.

340. Step 6 — Remove unsupported mechanism | R2.1

Submitted Discussion:

The intervention improved metacognition by clarifying next actions.

Revised:

Participant reports that the prompts clarified next actions are consistent with action clarity as one possible contributor to the benefit, but metacognition was not measured and the mechanism remains unresolved.

341. Propagation of R2.1

Remove metacognition from:

• Abstract conclusion;
• Discussion summary;
• Conclusion;
• graphical abstract if present.

342. Step 7 — Narrow population generalisation | R2.2

Submitted:

Structured feedback benefits bilingual learners.

Revised:

Structured feedback improved one-week independent revision among the advanced bilingual adult learners studied.

343. Propagation of R2.2

Title includes advanced bilingual learners.

Abstract includes adult sample.

Discussion adds high-support/single-site boundary.

Conclusion stops at studied sample.

344. Step 8 — Replace durable language | R2.3

Submitted title:

Structured Feedback Builds Lasting Independent Writers.

Revised title:

Structured Feedback Improves One-Week Independent Revision in Advanced Bilingual Learners: A Randomised Trial.

345. Why title moves to one-week despite 12-week data

The primary outcome is one week and the 12-week estimate is uncertain. The title should foreground the strongest evidence rather than the most ambitious horizon.

346. Step 9 — Add delayed-transfer literature | R2.4

Introduction:

field gap reframed more fairly: delayed evidence exists but is mixed and sparse in advanced learners.

Discussion:

12-week imprecision compared explicitly with prior weaker-persistence studies.

347. Contribution changes after literature correction

Original:

first delayed-transfer study.

Revised:

delayed unseen-task evidence in advanced bilingual adult learners.

Specific contribution replaces fragile priority claim.

348. Step 10 — Decline secondary-school replication | R2.5

Reason:

new population not available; not necessary to establish adult-sample result.

Scientific repair:

narrow generalisation and identify adolescent replication as next study.

349. Strong response to R2.5

We agree that replication in secondary-school learners would be valuable for assessing age-related transportability. However, the present study was designed and approved for advanced adult learners and we cannot collect a new adolescent sample within this revision. We have addressed the underlying generalisation concern by restricting all population-wide claims to the studied adult sample and adding adolescent replication as a future research priority (Discussion, lines …; Conclusion, lines …).

350. This is not “refusing the reviewer” | It solves the valid underlying concern

351. Revised paper identity after all changes

Object	Submitted	Revised
primary claim	lasting independent writing	one-week independent revision benefit
uncertainty	understated	cluster-adjusted + sensitivity
mechanism	metacognition asserted	unresolved; action clarity plausible
population	bilingual learners broadly	advanced bilingual adult sample
durability	claimed	12-week unresolved
subgroup	absent	exploratory

352. The revised paper is narrower but stronger | Successful major revision

It has:

• better uncertainty;
• more honest mechanism;
• clearer population scope;
• transparent exploratory analysis;
• fairer literature context;
• stronger treatment-effect wording where randomisation justifies it.

353. Revised Abstract | Example

We tested whether structured action-oriented feedback improves independent revision after feedback removal in 120 advanced bilingual adult learners. Participants were randomised to structured or standard evaluative feedback, and the primary outcome was an unseen revision task one week later. Assignment to structured feedback increased the one-week revision score by 5.7 points; cluster-adjusted uncertainty was wider than in the original analysis, and missing-data sensitivity produced a similar direction. The 12-week estimate was smaller and imprecise. These findings support a near-term independent-revision benefit in the studied sample but do not establish the active mechanism or longer-term durability.

354. Revised Discussion opening | Example

The revised analyses continue to support a near-term benefit of assignment to the structured-feedback condition on the primary one-week revision outcome, although the cluster-adjusted interval indicates greater uncertainty in magnitude than the original manuscript reported. The missing-data sensitivity analysis produced a similar direction. The smaller and imprecise 12-week estimate, however, leaves longer-term persistence unresolved.

355. Revised Conclusion | Example

Structured action-oriented feedback improved one-week independent revision relative to standard evaluative feedback among the advanced bilingual adults studied. The revised analyses support the direction of this near-term effect while indicating less precise magnitude than the submitted manuscript implied. The active mechanism, 12-week persistence and transportability beyond the high-support adult setting remain unresolved. Component-controlled and broader delayed replication are therefore the next critical tests.

356. Revision summary for the editor | Example

In response to the editor and reviewers, we made six major revisions: (1) re-estimated the primary uncertainty accounting for instructor clustering; (2) added multiple-imputation sensitivity analysis; (3) added a clearly labelled exploratory baseline-proficiency interaction analysis; (4) removed unsupported metacognition claims; (5) narrowed durability and population-generalisation language; and (6) expanded the delayed-transfer literature context. These changes narrow the manuscript’s central claim from durable writing development to a near-term independent-revision benefit in the studied advanced adult sample.

357. Point-by-point response R1.1 | Full example

Reviewer comment: “Participants were taught by six instructors. The analysis appears to treat observations as independent. Please account for clustering by instructor.”

Response: We agree that instructor-level clustering should be reflected in the uncertainty estimate. We re-estimated the primary baseline-adjusted regression with instructor-clustered standard errors. The point estimate remains 5.7 points, while the 95% confidence interval widens from 2.3–9.1 to 1.1–10.3. We have updated the statistical Methods, primary Results, Figure 2, Abstract and Discussion to reflect this increased uncertainty. The direction of the primary effect is unchanged, but we no longer describe the magnitude as precise.

358. Point-by-point response R1.2 | Full example

Reviewer comment: “Please assess sensitivity to missing-data assumptions.”

Response: We agree. We added a multiple-imputation sensitivity analysis under a missing-at-random assumption. The revised estimate was 5.3 points (95% CI 0.7–9.9), similar in direction to the complete-case analysis. We now describe the imputation model in Methods, report the sensitivity result in Results and Table S4, and revise the Discussion to state that the direction of the primary result did not materially depend on complete-case analysis.

359. Point-by-point response R1.3 | Full example

Reviewer comment: “The effect may be stronger for lower-proficiency learners. Please stratify by baseline proficiency.”

Response: We agree that baseline proficiency is a plausible candidate moderator. We added an exploratory group-by-baseline-proficiency interaction analysis. Because this analysis was not prespecified, we do not present it as confirmatory and we do not rewrite the Introduction as though the moderation hypothesis was planned. The interaction estimate is imprecise and is reported as hypothesis-generating in Results and Discussion, with full estimates in Figure S3.

360. Point-by-point response R1.4 | Full example

Reviewer comment: “The Conclusion could be stronger given randomisation.”

Response: We partly agree. Random assignment supports a causal effect of assignment to the structured-feedback condition on the measured one-week revision outcome, and we have strengthened this wording. We do not extend the stronger causal language to “writing ability” or the proposed mechanism because the study measured one revision task and did not isolate the intervention components. The revised Conclusion now distinguishes these levels explicitly.

361. Point-by-point response R2.1 | Full example

Reviewer comment: “Metacognition was not measured.”

Response: We agree. The submitted manuscript over-interpreted participant reports of clearer next actions as evidence of improved metacognition. We have removed the metacognition claim from the Abstract, Discussion and Conclusion. We now state that action clarity is one plausible contributor to the observed benefit but that the mechanism remains unresolved.

362. Point-by-point response R2.2 | Full example

Reviewer comment: “Claims about bilingual learners broadly are not justified.”

Response: We agree. The sample comprised advanced bilingual adults from one high-support programme. We now preserve this population boundary in the Title, Abstract, Discussion and Conclusion and identify replication in younger and lower-support settings as necessary before broader generalisation.

363. Point-by-point response R2.3 | Full example

Reviewer comment: “A 12-week follow-up does not establish durable writing ability.”

Response: We agree. We removed “lasting” and “durable writing ability” language throughout. We now distinguish the clear one-week primary effect from the smaller, imprecise 12-week estimate and state that longer-term durability remains unresolved.

364. Point-by-point response R2.4 | Full example

Reviewer comment: “The Discussion omits delayed-transfer studies with weaker persistence.”

Response: We thank the reviewer for identifying this omission. We added the relevant delayed-transfer literature to the Introduction and Discussion. This evidence also led us to revise our novelty claim: we no longer describe the study as the first delayed-transfer test and instead define the contribution more specifically as delayed unseen-task evidence in advanced bilingual adult learners.

365. Point-by-point response R2.5 | Full example

Reviewer comment: “Please repeat the study in secondary-school students.”

Response: We agree that adolescent replication would directly test age-related transportability. However, the current study and ethics approval concern advanced adult learners and we cannot collect a new adolescent sample within this revision. We have addressed the underlying scope concern by restricting all broader population claims and identifying adolescent replication as a separate future study. We therefore have not added the requested experiment to the current manuscript.

366. Why this response package works | Reviewer can audit the entire revision

Every comment has:

• a scientific position;
• a concrete change;
• appropriate temporal label;
• manuscript dependencies;
• updated final claim.

367. The response and manuscript tell the same story | Critical success criterion

Submitted story:

durable independent writing through metacognition.

Revised story:

near-term independent revision benefit; magnitude less precise; mechanism/durability/generalisation unresolved.

The response letter explicitly tells the editor this changed.

368. The authors did not “lose” the paper | They found its defensible centre

369. What if the editor wanted the secondary-school experiment anyway?

Then authors would need to decide whether:

• collecting the sample is feasible and ethical;
• the scope still fits one paper;
• an extension is possible;
• or withdrawal/resubmission elsewhere is more appropriate.

That is an editorial strategy question, not something to solve by pretending the adult study already generalises.

370. Part V operating rule | A successful revision may change the story

The goal of revision is not to preserve the submitted narrative. The goal is to produce the strongest manuscript that remains true after the reviewers’ best criticisms have been answered.

Revision 的目标不是保住原稿故事，而是让 paper 在 reviewer 最强的批评经过检验以后，留下一个更可靠、更透明、更守得住的 final story。

---

Part V completed a major-revision case. Part VI will turn the entire workflow into repeatable drills, bilingual response patterns, team controls, a 100-point revision rubric, final submission checklist and independent revise-and-resubmit benchmark.

Part VI — Revision operating system, bilingual drills and final resubmission gate | 第六部分：把 revise-and-resubmit 变成可重复系统

371. A revision team needs roles, not just tasks | Team control

Long revisions fail when five co-authors “help” independently. Assign explicit ownership:

• Revision lead: owns decision-letter map and final coherence.
• Analysis owner: owns reruns, new analyses and numerical source of truth.
• Methods/Results owner: ensures manuscript matches final analysis state.
• Claim owner: audits Abstract, Discussion and Conclusion after evidence changes.
• Response owner: keeps point-by-point document synchronized.
• Final verifier: checks every claimed change independently.

372. One person can hold several roles | But every role must exist

A two-author paper may divide these roles informally. A large collaboration should make them explicit.

373. Build one revision dashboard | Do not manage the revision from email threads

Comment ID	Owner	Decision	Scientific change	Dependencies	Response drafted	Verified
R1.1	statistics	accept	clustered uncertainty	M/R/F/A/D/C	yes	yes
R1.3	analysis	partly accept	exploratory interaction	M/R/D/S	yes	yes
R2.2	revision lead	accept	scope narrowed	T/A/D/C	yes	yes

374. “Done” and “verified” are different statuses

The person who makes a change should not be the only person deciding it is complete.

375. Comment closure rule | Close only after all dependencies are checked

R2.2 is not complete merely because one sentence in Discussion changed. It closes only after Title, Abstract, Discussion, Conclusion and response letter all reflect the narrower population scope.

376. Revision order by dependency | Do upstream science first

1. Data integrity and analysis corrections.
2. Sample/outcome definitions.
3. Methods and Results.
4. Figures and tables.
5. Discussion and limitations.
6. Conclusion.
7. Introduction.
8. Abstract.
9. Title.
10. Response document.
11. Language polish.

377. Why Introduction can come late in revision

If analyses or claims changed, the Introduction may need to be re-scoped after the revised evidence state is stable. Otherwise authors can waste time repeatedly rewriting the gap.

378. Why style comes last

Copyediting an unstable paragraph is wasted work. Worse, stylistic synonym replacement can reintroduce construct drift after scientific terms were fixed.

379. The 10-minute revision control scan | Before every team meeting

1. Which comments remain open?
2. Which scientific objects changed since last meeting?
3. Which dependencies are not propagated?
4. Did any new analysis alter the primary claim?
5. Does the response document still match the manuscript?

380. The 20-minute response audit

Time	Task
4 min	check every editor comment has a response
4 min	check every reviewer comment has a response
4 min	check every “we changed” statement against manuscript
4 min	check every “no change” response has scientific rationale
4 min	check line/section/table references

381. The 45-minute revised-manuscript audit

Time	Task
8 min	rebuild revised identity card
8 min	compare submitted vs revised primary claim
8 min	reconcile all new/changed numbers
7 min	audit planned vs reviewer-added analyses
7 min	audit limitations/scope propagation
7 min	audit Title/Abstract/Conclusion agreement

382. The 90-minute final resubmission audit

Time	Task
10 min	decision letter and editor priorities
10 min	response matrix completeness
15 min	new analysis/data verification
10 min	Methods–Results consistency
10 min	figures/tables/supplement consistency
10 min	claim/scope/causality audit
10 min	response–manuscript cross-check
10 min	bilingual terminology/tone audit
5 min	submission file checklist

383. Multi-round revision | R2 is not “do R1 again”

Round 2 begins from the revised manuscript, not from the original submission.

384. Build a round-to-round state table

Object	Submission	R1	R2
primary claim	durable benefit	near-term benefit	near-term benefit
primary model	ordinary SE	clustered SE	clustered SE
subgroup	none	exploratory	exploratory
mechanism	metacognition claimed	unresolved	unresolved

385. Round 2 reviewer may challenge a Round 1 change

Do not pretend the new element was original. Explain the revision history where it matters.

386. Example: R2 reviewer questions exploratory subgroup added in R1

Response:

This subgroup analysis was added during the first revision at Reviewer 1’s suggestion and was explicitly labelled exploratory. We agree with the present reviewer that the estimate is too imprecise for a substantive moderation conclusion, and we have further reduced its emphasis in the Discussion.

387. Example: R2 reviewer asks to make exploratory subgroup primary

Do not rewrite history. You can state that the subgroup is hypothesis-generating and propose prospective confirmation.

388. Example: editor asks authors to reconcile reviewers rather than choose one

Write the scientific rationale that resolves the apparent conflict.

389. Bilingual revision risk | Internal Chinese notes can make responses sound too strong

Chinese team note:

审稿人这个意见不对，我们坚持原分析。

Risky English:

The reviewer is incorrect and we insist on our original analysis.

Professional reconstruction:

We respectfully retain the prespecified analysis because the requested alternative conditions on a post-treatment variable and would estimate a different causal quantity. We have added this rationale to Methods and provide an additional sensitivity analysis to address the underlying concern.

390. Chinese “我们已经修改” | English needs what + where + why

Weak:

We have revised this.

Strong:

We now define the primary outcome as the one-week unseen revision score in both the Abstract and Methods and use the same term throughout Results and Discussion.

391. Chinese “感谢审稿人的宝贵意见” | Do not repeat formula mechanically

A brief thanks at the start is enough. Repeating an elaborate gratitude sentence before every comment can make the response longer without making it clearer.

392. Chinese “我们完全同意” | Use only when you fully agree

If you agree with concern but not remedy, write:

We agree with the underlying concern but have addressed it differently for the following reason…

393. Chinese “我们认为” | Evidence is stronger than opinion framing

Weak:

We think the current method is better.

Strong:

The current method preserves the prespecified estimand, whereas the requested adjustment would condition on a post-treatment variable.

394. Chinese “由于篇幅限制” | Space alone is weak scientific justification

If material is moved to Supplement, explain that core evidence remains in main text and technical detail is relocated to meet format constraints.

395. Chinese “由于条件限制无法完成” | Explain the scientific consequence

We cannot collect the requested six-month outcome retrospectively. We have therefore removed the durability claim and now identify longer follow-up as the unresolved question.

396. Chinese “审稿人可能误解了” | Own ambiguity where appropriate

We recognise that our original wording could reasonably be interpreted this way and have revised it for clarity.

397. Chinese “我们保留原文” | No-change response still needs rationale

We retain the original term because it is the validated construct measured by the instrument. We have added its definition at first use to avoid ambiguity.

398. Chinese “这个分析没有必要” | Explain why

The requested analysis does not answer the prespecified question and would be underpowered in the available subgroup. We therefore do not present it as an inferential analysis.

399. Chinese “结果没有变化” | Specify what stayed stable

The point estimate and direction were similar, while the confidence interval widened modestly.

400. Chinese “结论不受影响” | State which conclusion

The revised analysis does not change the primary conclusion that the structured condition improved the one-week measured outcome, although the magnitude is less precise.

401. Chinese “进一步验证了” | Revision-added analysis may not “validate”

Use:

The sensitivity analysis produced a similar estimate, reducing concern that the result depends on this analytic assumption.

402. Chinese “补充实验” | New experiment must be integrated, not appended

If a new experiment changes mechanism or scope, restructure the manuscript accordingly.

403. Chinese “根据审稿人建议” | Good in response, not always necessary in published manuscript

The response document should preserve revision chronology. The final published manuscript should label analyses scientifically—such as exploratory/post hoc—without necessarily narrating reviewer identity.

404. Chinese “显著改善” after revision | Re-check statistical and practical meaning

If new analysis widens uncertainty, do not retain a stronger old adjective.

405. Bilingual response lexicon | Useful distinctions

Chinese working concept	English response move
同意意见	We agree that…
同意问题，不同意方法	We agree with the concern but address it by…
礼貌不同意	We respectfully disagree because…
无法完成	We cannot perform X because…; we therefore…
澄清误解	We recognise the original wording was ambiguous and now clarify…
新增探索分析	We added an exploratory analysis in response to…
不改正文	No manuscript change was made because…

406. Drill 1 — Comment classification

Reviewer:

“Your title claims causality, but the study is cross-sectional.”

Classify and respond.

407. Model answer 1

Type: interpretation/scope/causal validity. Accept. Narrow title and all causal claims to association language; audit Abstract, Discussion, Conclusion and graphical abstract.

408. Drill 2 — Concern valid, remedy poor

Reviewer:

“Baseline imbalance is a problem; therefore delete half the sample until groups match.”

409. Model answer 2

Partly accept the concern, reject arbitrary deletion. Use a defensible adjustment/sensitivity strategy appropriate to design and explain why.

410. Drill 3 — Reviewer requests post-hoc subgroup

What must not change?

411. Model answer 3

Do not rewrite Introduction as if subgroup was hypothesised. Label analysis exploratory/post hoc. Use interaction rather than separate significance logic where appropriate. Keep primary conclusion primary.

412. Drill 4 — Reviewer asks for new child sample

Adult study. What are the two legitimate paths?

413. Model answer 4

Collect new sample if feasible and scientifically central, integrating it as a genuine new evidence node; or decline, narrow population claims and identify child/adolescent replication as future work.

414. Drill 5 — New analysis weakens significance

What must propagate?

415. Model answer 5

Update Results, figures/tables, Abstract, Discussion and Conclusion; remove old “robust/significant/precise” language where no longer accurate.

416. Drill 6 — Reviewer misreads comparator

Control did receive feedback, but Methods were unclear.

417. Model answer 6

Correct the factual misunderstanding respectfully and revise Methods/Abstract/figure labels so the comparator is impossible to miss.

418. Drill 7 — Reviewer asks to remove null primary outcome

419. Model answer 7

Respectfully disagree. Retain prespecified primary result and report estimate/uncertainty clearly; removing it would distort evidence hierarchy.

420. Drill 8 — Reviewer asks to add own irrelevant citation

421. Model answer 8

Assess relevance objectively. If not relevant, politely decline and explain that it does not address the specific evidence gap.

422. Drill 9 — Two reviewers conflict on Discussion length

423. Model answer 9

Remove repetitive Results and low-value background, then use saved space for the high-value limitation/theory comparison requested by the other reviewer.

424. Drill 10 — Reviewer asks to call a four-week effect “long-term”

425. Model answer 10

Do not comply unless field conventions justify it. Use the exact four-week horizon and explain longer-term durability remains unknown.

426. Drill 11 — New experiment contradicts original effect

427. Model answer 11

Report both. Reassess primary narrative, identify boundary conditions or inconsistency, reduce certainty and update Abstract/Discussion/Conclusion. Do not hide revision-added contradictory evidence.

428. Drill 12 — Reviewer asks for mechanism you cannot test

429. Model answer 12

Remove mechanism certainty, present mechanism as a hypothesis, explain inability to collect required evidence and identify the test as future work.

430. Drill 13 — Reviewer asks for stronger treatment-effect language in randomised study

431. Model answer 13

Strengthen causal wording only for assignment to the full condition on the measured outcome. Do not automatically strengthen mechanism, construct breadth or external validity.

432. Drill 14 — Reviewer asks for weaker causal language everywhere despite randomisation

433. Model answer 14

Respectfully defend causal treatment-effect language while clarifying the non-causal dimensions—mechanism, component attribution and generalisation.

434. Drill 15 — Response says change made, manuscript still old

435. Model answer 15

Do not submit. Reconcile response and manuscript; this is a credibility failure.

436. Drill 16 — Round 2 reviewer asks to reverse Round 1 analysis

437. Model answer 16

Explain the analysis history and scientific rationale. Consider sensitivity analyses or editorial clarification rather than silently oscillating.

438. Drill 17 — Reviewer comment is rude but scientifically correct

439. Model answer 17

Ignore tone in reviewer-facing response; address the scientific concern fully. Raise conduct issue separately with editor only if necessary.

440. Drill 18 — Reviewer comment is polite but scientifically invalid

441. Model answer 18

Respectfully disagree with evidence/method rationale. Do not adopt an invalid change merely because tone is friendly.

442. Drill 19 — Reviewer says “make paper more impactful”

443. Model answer 19

Ask what contribution is actually supported. Clarify significance through specific contribution, not inflated certainty or scope.

444. Drill 20 — Editor requires shorter response letter

445. Model answer 20

Compress repeated rationale, cross-reference related analyses and keep each comment’s decision/change/location intact. Do not omit substantive responses.

446. Seven-day revision training cycle | 七天训练

1. Day 1: classify 30 reviewer comments by request type.
2. Day 2: write accept/partly accept/disagree/cannot-address responses.
3. Day 3: build dependency graphs for 10 scientific changes.
4. Day 4: repair response–manuscript mismatches.
5. Day 5: handle conflicting reviewers and editor priorities.
6. Day 6: bilingual response reconstruction.
7. Day 7: complete fictional major revision.

447. Twelve-week C1–C2 revise-and-resubmit progression

Weeks	Focus	Output
1–2	comment classification + tone	response library
3–4	scientific disagreement + alternatives	rebuttal portfolio
5–6	analysis/experiment dependency propagation	change logs
7–8	whole-manuscript revision control	revised manuscripts
9–10	multi-round/editor conflict handling	R1/R2 case files
11–12	independent full resubmission	clean + tracked + response package

448. The 100-point peer-review revision rubric

Dimension	Points	Mastery evidence
Decision-letter control	10	editor priorities, deadline and file requirements are explicit
Comment classification	10	concern type and proposed remedy distinguished
Scientific validity	15	valid concerns resolved; invalid changes resisted with evidence
Temporal honesty	10	reviewer-added/post-hoc analyses remain labelled
Dependency propagation	15	all affected sections/displays updated consistently
Response-letter completeness	10	every comment answered with decision/change/location
Disagreement quality	10	respectful, evidence-based, solution-oriented
Response–manuscript alignment	10	every claimed revision exists and every major change is documented
Version/team control	5	submitted baseline, clean/tracked versions and analysis state preserved
Bilingual precision	5	Chinese internal notes do not create defensive or overstrong English

449. What 90–100 looks like | C2-ready revise-and-resubmit

The revised paper is scientifically stronger than the submitted paper, not merely longer. Every new analysis has a purpose and temporal label. Every accepted change propagates cleanly. Every disagreement is principled. Reviewers can verify changes quickly. The response and manuscript describe the same revised evidence state.

450. What 75–89 looks like | Strong C1 revision

Most comments are handled well, but one or two local problems remain: perhaps an exploratory analysis is over-emphasised, a changed limitation has not reached the Abstract, or a response says “changed” without a precise location.

451. What 60–74 looks like | Polite responses, unstable science

The response letter sounds professional but changes are treated comment by comment. New analyses are added without propagation, manuscript history is blurred, and sections begin to contradict each other.

452. Below 60 | Comment compliance without revision control

The team is chasing reviewers, not controlling a scientific argument. Rebuild the revision matrix and identity card before further prose editing.

453. Final resubmission package checklist | Check journal-specific requirements too

1. Clean revised manuscript.
2. Tracked/marked manuscript if requested.
3. Point-by-point response.
4. Cover letter if required.
5. Updated figures.
6. Updated tables.
7. Updated supplementary files.
8. Updated reporting checklist if applicable.
9. Updated data/code availability.
10. Updated ethics/disclosures if needed.
11. Final line/page references.
12. Consistent file names.

454. Scientific final gate | Before uploading files

Ask:

1. What is the revised primary claim?
2. Did any new evidence strengthen or weaken it?
3. Does the Title match it?
4. Does the Abstract match it?
5. Does the Discussion match it?
6. Does the Conclusion match it?
7. Are all reviewer-added analyses labelled honestly?
8. Do all numbers come from the final analysis state?
9. Does the response letter describe this exact manuscript?

455. Reviewer-comment coverage gate

Search the response file for every comment ID. None should be missing.

456. “We changed” gate

Search for every phrase such as:

we added / revised / removed / reanalysed / clarified / updated.

Verify each against the manuscript.

457. “No change” gate

Search for:

we retain / we did not / no change.

Verify each has a scientific reason.

458. New-analysis gate

List every analysis that did not exist at submission. Confirm planned/exploratory status and manuscript location.

459. New-data gate

List every new experiment or dataset and confirm integration into Methods, Results, Discussion, Abstract and response as appropriate.

460. Deleted-claim gate

List every important claim removed during revision. Ensure no old version survives in Title, Abstract, figures, highlights or Conclusion.

461. Bilingual final gate

Translate the revised primary claim into natural Simplified Chinese and back into English. Check whether:

• causality changed;
• scope widened;
• certainty increased;
• “effective” replaced a specific outcome;
• “long-term” appeared without the measured horizon.

462. Final independent benchmark | Revise a paper you did not write

Use a fictional paper or an openly available manuscript-review pair. Complete:

1. extract editor priorities;
2. classify every reviewer comment;
3. make accept/partial/disagree/cannot decisions;
4. build scientific change log;
5. propagate five major changes;
6. write response summary;
7. write ten point-by-point responses;
8. build revised identity card;
9. audit Title/Abstract/Discussion/Conclusion;
10. score with the 100-point rubric.

463. Final assignment | Your own revise-and-resubmit package

Take one manuscript and one realistic reviewer set. Produce:

1. decision-letter map;
2. comment-ID matrix;
3. revision triage;
4. dependency graph for every major change;
5. submitted and revised identity cards;
6. scientific change memo;
7. clean revised manuscript;
8. tracked/marked version;
9. point-by-point response;
10. editor cover letter;
11. final 100-point audit.

464. Research and reference floor | 研究与参考基础

• Nature Computational Science | How to Respond to Reviewers — revision summary, full reviewer comments, point-by-point responses, disagreement and editor-facing concerns.
• Nature | Revisions and Feedback — re-review of revised manuscripts and reviewer assessment of revisions.
• Nature Geoscience | The Art of Responding to Reviews — clarity, tone, following changes through the manuscript and standing up for sound science.
• Springer Nature | Revising and Responding — address all points, explain major revisions and justify changes not made.
• Elsevier | Peer Review and Resubmission — objective evidence-based responses and disagreement handling.
• Elsevier | Responding to Reviewer Comments the CALM Way — comment-by-comment response workflow and revision files.
• Nature Communications | Editorial Process — revised manuscript, cover letter and point-by-point response.
• Scientific Reports | Peer Review — revised submissions and response to reviewers.

465. Canonical eduKate route | Previous research-writing lessons

• Lesson 020 | Discussion Without Overclaiming
• Lesson 021 | Conclusion Closure
• Lesson 022 | Full Research-Paper Coherence

466. SEO language map | 本课自然覆盖的搜索意图

This lesson naturally serves readers searching for response to reviewers, peer review response letter, revise and resubmit, major revision, minor revision, rebuttal letter, point-by-point response, how to disagree with reviewer, reviewer comments response examples, revise manuscript after peer review, additional analysis after peer review, post hoc analysis reviewer request, response to editor, tracked changes manuscript, revision cover letter, manuscript revision workflow, academic English peer review, C1 academic writing, C2 academic English, English for Mandarin speakers, 审稿意见回复, 回复审稿人, 大修, 小修, 修改稿, rebuttal 怎么写, peer review 回复, 学术英语修稿 and 中文母语学术英语.

467. Final quality gate | 24 questions before resubmission

1. Did we read the editor’s letter as the controlling decision document?
2. Does every editor/reviewer comment have an ID and response?
3. Did we distinguish the concern from the reviewer’s proposed solution?
4. Were validity-critical concerns prioritised?
5. Are invalid requests respectfully resisted?
6. Are impossible requests paired with scientific claim repair?
7. Are all reviewer-added analyses labelled according to timing?
8. Did any new analysis change the primary estimate or uncertainty?
9. Were those changes propagated everywhere?
10. Did any new experiment alter the paper identity?
11. Were all figures/tables regenerated from the final analysis state?
12. Are sample counts and denominators reconciled?
13. Are revised limitations reflected in scope and conclusion?
14. Did any literature addition change the novelty/contribution claim?
15. Do Title, Abstract, Discussion and Conclusion describe the same revised paper?
16. Does every “we changed” statement match the manuscript?
17. Does every “no change” statement contain scientific rationale?
18. Are conflicting reviewer requests resolved coherently?
19. Are confidential concerns kept in editor-only communication?
20. Are response locations/line references final?
21. Are clean and tracked versions based on the same manuscript state?
22. Is the submitted version preserved as an archive?
23. Did bilingual shorthand introduce stronger or more defensive English?
24. Is the revised paper scientifically better than the submitted one?

468. Final principle | Peer review is an adversarial test of coherence

A strong revise-and-resubmit does not preserve every sentence of the original paper and does not obey every reviewer request. It preserves something more important: a traceable, evidence-controlled research identity that becomes clearer after criticism.

Reviewer comments are inputs.

Scientific judgement decides what changes.

Dependency control keeps the manuscript whole.

The response letter makes the changes auditable.

好的 revision，不是“原稿尽量不动”，也不是“reviewer 说什么都做”。真正要保护的是：经过批评以后，整篇 paper 的 research identity 仍然 traceable、evidence-controlled，而且比以前更清楚。

469. Exit standard | You are ready to move on when…

You can receive a difficult review set and, without becoming defensive or compliant by reflex:

• classify every concern;
• identify the underlying scientific object;
• choose accept/partial/disagree/cannot honestly;
• propagate accepted changes through the manuscript;
• preserve planned-vs-exploratory history;
• write a precise point-by-point response;
• handle reviewer conflict;
• audit the final resubmission package independently.

At that point, peer review becomes controlled scientific revision rather than a stressful list of comments.

---

Next lesson reserved | 下一课

EDKS-ADV-ZH-0024 · Lesson No.024 · Write a Peer Review Report That Improves the Paper Without Rewriting It as Your Own | 写好 Peer Review：严格、具体、可执行，但不要把别人的论文改成自己的

The next lesson will reverse the perspective. It will teach how to review another manuscript fairly: identify validity-critical issues, separate major from minor concerns, avoid scope creep, write actionable comments, calibrate recommendation language, distinguish preference from error and produce a report that helps editor and author without turning the manuscript into the reviewer’s own paper.

Back to Advanced English Chinese Edition Hub · 返回高级英语中文版主页