Series ID: EDKS-ADV-ZH-0022 · How to Learn Advanced English (Chinese Edition) · Lesson No.022 · C1 → C2

Write a Full Research Paper Whose Sections Agree With One Another | 让整篇研究论文各部分彼此一致

A research paper can contain excellent sentences, accurate statistics and individually respectable sections—and still fail because the sections are answering different versions of the study.

一篇 research paper 可以每一个 section 单独看都写得不错，却整体失去可信度。最常见的原因不是 grammar，而是 paper 在不同 section 里悄悄变成了不同的研究：Introduction 问 A，Methods 测 B，Results 报 C，Discussion 解释 D，Conclusion 最后又宣布 E。

At C1–C2 level, full-paper quality therefore requires more than section-level writing skill. It requires cross-section coherence: every major claim must have a traceable lineage from the question that motivated it, through the method that made it observable, to the result that reported it, the discussion that interpreted it and the conclusion that finally compressed it.

The governing chain is:

QUESTION → CONSTRUCT → DESIGN → MEASURE → ANALYSIS → RESULT → INTERPRETATION → CONCLUSION.

问题 → construct → design → measure → analysis → result → interpretation → conclusion。

If one link changes silently, the paper develops research identity drift.

---

Part I — The paper-level contract | 第一部分：整篇论文必须遵守同一个 contract

1. Why this lesson exists after Lessons 016–021 | Section specialists are not enough

The preceding lessons own individual research-writing jobs:

• Lesson 016 — Abstract compression.
• Lesson 017 — Introduction architecture.
• Lesson 018 — Methods transparency.
• Lesson 019 — Results reporting.
• Lesson 020 — Discussion without overclaiming.
• Lesson 021 — Conclusion closure.

Lesson 022 owns the integration problem: do all those sections describe the same study?

2. Nature: Methods are the backbone because findings depend on design | The paper is one inferential system

Nature Cell Biology describes Methods as the backbone of a research paper because findings stand or fall with design, analysis, statistical rigour and reproducibility. Reviewers examine methods and presented data together to judge whether conclusions are supported and whether gaps in reasoning remain.

Nature Cell Biology | Methodical about Methods

This is paper-level coherence: conclusion quality cannot be separated from design quality, and design meaning cannot be separated from the question it was built to answer.

3. CONSORT 2025: readers should not infer silent changes | Tell readers what was actually planned and done

CONSORT 2025 emphasises clear, complete and transparent reporting and specifically highlights alignment among protocol, design, conduct, analysis and final reporting. It notes that readers should be told explicitly what was done rather than forced to infer it, and that changes from protocol or analysis plans should be acknowledged.

CONSORT 2025 Statement

Although CONSORT applies to randomised trials, the underlying integrity principle generalises:

planned question → actual method → actual analysis → reported result must remain traceable.

4. Reporting guidelines are consistency maps | Checklists reveal missing links

The EQUATOR Network organises reporting guidelines by study design because different evidence types require different information to remain interpretable. A checklist is not a substitute for reasoning, but it can expose when a manuscript contains a claim without the method, result or context needed to support it.

EQUATOR Network | Reporting Guidelines

5. Modern AI research makes whole-paper alignment even more visible | TRIPOD-LLM spans every section

TRIPOD-LLM organises reporting items across title, abstract, introduction, methods, results, discussion and other manuscript components. That section-spanning structure reflects a deeper principle: transparency is distributed across the entire paper, not confined to one “Methods” paragraph.

Nature Medicine | TRIPOD-LLM Reporting Guideline

6. The paper-level contract has ten invariants | 十个 invariant

A coherent paper tries to preserve ten identities across sections:

1. Research question identity.
2. Construct identity.
3. Population identity.
4. Sample identity.
5. Exposure/intervention identity.
6. Outcome identity.
7. Time-point identity.
8. Analysis identity.
9. Claim-strength identity.
10. Contribution/scope identity.

7. Research question identity | The question should not mutate

Introduction:

Does structured feedback improve independent revision at one week?

Results:

self-reported confidence improved.

Conclusion:

Structured feedback improves learner independence.

This paper has drifted from measured revision to confidence to a broad independence construct.

8. The exact question can become more precise—but not silently different | Refinement vs mutation

Refinement:

Does X improve Y? → Does X improve one-week Y relative to standard practice?

Mutation:

Does X improve Y? → Do participants like X?

9. Construct identity | Keep “what you mean” stable

Construct drift occurs when the label becomes broader or different:

• revision quality → writing ability;
• attendance → engagement;
• confidence → competence;
• benchmark score → intelligence;
• clicks → successful use;
• symptom scale → overall health.

10. Operational measure and conceptual construct are allowed to differ—but the relationship must be explicit

Introduction may define independent revision capability.

Methods operationalises it as score on an unseen task without feedback.

Discussion can infer back toward the construct, but must acknowledge what the measure does and does not cover.

11. Population identity | Who is the paper about?

Title:

Feedback in Advanced Bilingual Learners.

Methods:

adult volunteers in one programme.

Conclusion:

students benefit.

The noun has expanded.

12. Population can narrow due to eligibility or attrition | Report the actual inference set

Target population, eligible population, enrolled sample, analysed sample and final follow-up sample may differ. The paper should show those transitions.

13. Sample identity | n must reconcile

Abstract says n = 120.

Methods says 118 randomised.

Figure says n = 104.

Results says 109 analysed.

These numbers can all be correct—but only if participant flow explains why.

14. Denominator drift is a major coherence failure | Percentages need their n

If 75% means 75/100 in one table and 75% means 9/12 in another, the paper should make denominators visible.

15. Intervention/exposure identity | What exactly is X?

Introduction:

structured feedback.

Methods:

structured prompts + extra planning time + longer feedback.

Discussion:

prompt structure caused the effect.

The intervention identity contains bundled components that the conclusion has separated without evidence.

16. Comparator identity matters too | Compared with what?

“Improved” means something only relative to a comparator or baseline. The comparator should remain consistent in abstract, Results and Conclusion.

17. Outcome identity | Primary outcome cannot silently change

Protocol/Introduction:

independent revision score.

Results:

primary result null; confidence positive.

Abstract/Conclusion:

intervention was effective.

This is outcome switching or narrative hierarchy drift.

18. Outcome hierarchy must survive publication | Primary remains primary

Secondary outcomes may be important, but they should remain labelled and proportionate.

19. Time-point identity | Immediate vs delayed is not cosmetic

A result at one week and a result at 12 weeks answer different questions. Conclusions about durability require delayed evidence.

20. “Post-intervention” can hide multiple horizons | Use exact timing when interpretation depends on it

21. Analysis identity | The analysis described should produce the result reported

Methods:

linear regression adjusted for baseline.

Results:

unadjusted group means presented as primary effect.

Not necessarily wrong—but the paper should distinguish descriptives from primary model estimates.

22. Analysis-plan changes need traceability | Planned vs post-hoc

If the primary model changed after inspecting data, report the change and reason. Do not rewrite the manuscript as though the final analysis was always the only plan.

23. Claim-strength identity | Certainty should not grow section by section

Abstract:

was associated with.

Discussion:

may contribute to.

Conclusion:

causes.

This is causal drift.

24. Certainty can legitimately change if analysis changes evidence | But explain why

For example, sensitivity analyses may increase confidence that the primary direction does not depend on one analytic choice. That can justify stronger confidence, but not a different causal design.

25. Contribution identity | What is the paper actually adding?

Introduction gap:

no delayed independent measure.

Conclusion contribution:

new theory of learner autonomy.

The contribution has shifted away from the stated problem.

26. Scope identity | The paper should leave with the same world it entered

A local study may discuss broader principles, but empirical scope must remain distinguishable from theoretical implication.

27. The alignment table | One row per invariant

Invariant	Introduction	Methods	Results	Discussion	Conclusion
question	what is unknown	design answers it	answer reported	answer interpreted	answer compressed
construct	defined	operationalised	measured	inferred carefully	scope preserved
population	target	sampled	analysed	generalised cautiously	boundary retained
outcome	priority	measurement	estimate	meaning	final claim

28. This table is a diagnostic, not a writing template | Use before submission

The paper can have different prose in every section while preserving the same invariant.

29. Title belongs in the contract | Title is not marketing outside the paper

The title should not be more causal, broader or more durable than the body.

30. Abstract belongs in the contract | Front-door coherence

The abstract should accurately compress the full paper, including outcome hierarchy and major uncertainty.

31. Tables and figures belong in the contract | Visual evidence is part of the manuscript

Labels, sample sizes, time points and outcome names should agree with the prose.

32. Supplementary material belongs in the contract | It cannot contradict the main text

A primary outcome cannot disappear into Supplementary Table S7 while the main text foregrounds a secondary positive result.

33. Protocol/registration belongs in the contract where applicable | Pre-publication identity

Protocol, trial registry and statistical analysis plan can reveal whether outcomes, analyses or hypotheses changed.

34. Data/code availability can extend the contract | Executable traceability

For computational work, code and data can make analysis lineage directly inspectable.

35. Ethics/consent language must agree with participant description | Same study population

Eligibility, recruitment and ethics statements should not describe different populations.

36. Terminology identity | One concept, one stable name where possible

If you alternate among:

revision transfer / writing independence / autonomous learning / writing ability

the reader may assume these are equivalent. If they are not, define the differences.

37. Synonym variation is not always elegant in technical prose | Consistency can beat stylistic variety

In research writing, repeating the precise technical term is often safer than replacing it with a near-synonym that changes meaning.

38. Variable-name identity | X must remain X

Methods:

baseline proficiency.

Results:

initial ability.

Discussion:

prior knowledge.

These may not be identical constructs.

39. Group-label identity | Avoid condition renaming

structured feedback group should not become high-support learners unless support truly defines the condition.

40. Time-label identity | Week 1, immediate post-test, T2—make mapping obvious

41. Figure-number identity | Every referenced display must exist and match content

42. Statistical-test identity | Test named in Methods should match Results

43. Unit identity | ms vs s, mg vs g, percentage vs percentage points

Unit drift can create large factual errors even when prose is fluent.

44. Scale-direction identity | Higher score = better or worse?

If one scale reverses direction, say so. Discussion should not interpret score increases incorrectly.

45. Missing-data identity | Methods plan vs actual missingness vs sensitivity

Methods:

how missing data will be handled.

Results:

how much data were missing and what analyses resulted.

Discussion:

what missingness does to confidence.

46. Exclusion identity | Who was excluded and when?

Pre-specified exclusion in Methods should reconcile with participant counts in Results.

47. Version identity | Instrument/software/model versions matter

Especially for rapidly changing computational systems, model/software version can be part of study identity.

48. Source-data identity | Same dataset throughout

If analyses use different subsets, tables and text should state which subset supports which claim.

49. Hypothesis identity | Pre-specified vs exploratory

Introduction should not retrospectively frame an exploratory finding as the central original hypothesis.

50. Research-paper coherence is temporal honesty | Preserve when decisions happened

Readers need to know which questions and analyses existed before results were known and which emerged afterwards.

51. HARKing is cross-section drift | Hypothesis After Results Are Known

When unexpected Results are rewritten into Introduction as though predicted, the paper’s temporal identity changes.

52. Outcome switching is cross-section drift | Primary becomes secondary after data

Transparency requires explicit reporting of changes where applicable.

53. Spin is often cross-section hierarchy drift | Numbers may be correct

A manuscript can report all data correctly yet foreground different evidence in Abstract/Conclusion than in the prespecified primary analysis.

54. Coherence is not rigidity | Papers can learn from data

Exploratory findings, revised models and unexpected interpretations are legitimate. The requirement is labelled change, not frozen thinking.

55. The labelled-change principle | If identity changes, show the transition

Pre-specified primary analysis…

Exploratory post-hoc analysis…

Protocol deviation…

Revised hypothesis generated by…

56. Readers can tolerate complexity better than hidden inconsistency | Transparency beats fake simplicity

57. The one-paper test | Could you describe the study in one stable sentence?

Template:

In [population], we used [design] to test whether [X] changes/is associated with [Y] at [time], using [primary analysis].

Every section should remain compatible with that sentence.

58. Example stable study sentence

In advanced bilingual adult learners, we used a randomised two-group trial to test whether structured feedback improves one-week independent revision relative to standard feedback, using baseline-adjusted regression as the primary analysis.

59. The paper identity card | Build before drafting

Field	Study identity
Population	advanced bilingual adult learners
Design	parallel randomised trial
Exposure/intervention	structured action-oriented feedback
Comparator	standard evaluative feedback
Primary outcome	one-week unseen revision score
Primary analysis	baseline-adjusted linear regression
Main scope	near-term independent revision

60. Part I operating rule | 整篇 paper 只有一个 research identity

A coherent research paper does not force every section to repeat the same sentence. It forces every section to remain accountable to the same study identity.

一致性不是每个 section 复制同一句话，而是每个 section 都对同一个 study identity 负责。

---

Part I defines the cross-section contract. Part II will follow the paper section by section, showing exactly what must pass forward from Title and Abstract into Introduction, Methods, Results, Discussion and Conclusion without mutation.

Part II — Section handoffs: what each part must receive and pass forward | 第二部分：每个 section 的输入与输出必须对得上

61. Think of the manuscript as a relay, not a stack | 论文不是六个独立盒子

Each section receives an intellectual object from the previous section, transforms it for a new purpose and hands it forward. A section can be beautifully written yet damage the paper if it hands forward a different object.

The relay is:

Title/Abstract orient → Introduction defines the problem → Methods operationalises it → Results reports the evidence → Discussion interprets the evidence → Conclusion closes the exact problem.

62. Title input contract | Title receives the final study identity

The title should inherit:

• the actual population if central;
• the actual intervention/exposure;
• the actual outcome/phenomenon;
• the actual design when important to interpretation;
• the correct causal strength.

63. Title output contract | Title gives the reader a truthful first model

A strong title creates expectations that the body later fulfils.

Weak title:

Structured Feedback Builds Independent Writers.

Study:

one-week revision outcome in advanced adults.

Stronger:

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

64. Title drift type 1 — causal inflation

Body:

associated with.

Title:

causes / improves / prevents.

Repair by matching design.

65. Title drift type 2 — construct inflation

Measured:

revision score.

Title:

writing ability.

Repair by naming the actual outcome or a justified construct.

66. Title drift type 3 — duration inflation

Measured:

four weeks.

Title:

lasting / durable / long-term.

67. Title drift type 4 — population inflation

Sample:

advanced adult volunteers.

Title:

students / learners generally.

68. Abstract input contract | Abstract receives the whole completed paper

The Abstract should be written from the final evidence state, not from the pre-analysis hopes that existed when the study began.

69. Abstract output contract | Abstract compresses without changing hierarchy

It should preserve:

• problem/question;
• design;
• primary outcome;
• main result;
• major uncertainty/boundary where material;
• final conclusion.

70. Abstract–body mismatch is high-impact | Many readers see only the Abstract

If the Abstract overclaims, the paper’s most visible surface misrepresents the full work even if the Discussion is careful.

71. Abstract drift type 1 — primary outcome disappears

Primary performance outcome null.

Abstract highlights secondary confidence gain.

This changes the paper’s evidence hierarchy.

72. Abstract drift type 2 — Methods are simplified until design meaning changes

Full paper:

non-randomised prospective cohort.

Abstract:

intervention study.

The shorter label may imply stronger causal control.

73. Abstract drift type 3 — uncertainty removed

Full paper:

estimate imprecise.

Abstract:

effective.

74. Abstract drift type 4 — subgroup promoted to headline result

Exploratory subgroup should remain exploratory.

75. Abstract audit question | Could someone reproduce the paper’s evidence hierarchy from the abstract?

They may not know every detail, but they should know which result is primary and which claims remain uncertain.

76. Introduction input contract | Introduction receives the current field state

It should accurately represent:

• what is already known;
• what is genuinely unresolved;
• why the unresolved state matters;
• what exact question the study addresses.

77. Introduction output contract | It hands Methods a researchable question

The final Introduction paragraph should be specific enough that the reader can predict what the Methods must contain.

If the question is:

Does structured feedback improve one-week independent revision?

Methods must contain:

• structured feedback definition;
• comparison condition;
• one-week timing;
• independent revision measure;
• analysis of the comparison.

78. If Methods cannot answer the Introduction question, the problem is upstream | Question–design mismatch

Question:

Does X cause Y?

Design:

cross-sectional correlation.

Either revise the question or justify a causal identification strategy.

79. Introduction drift type 1 — gap larger than study

Introduction:

How can we solve global learner disengagement?

Study:

one classroom survey.

The research space is too large for the evidence.

80. Introduction drift type 2 — outcome not actually measured

Introduction promises:

long-term learning.

Methods measure:

immediate task performance.

81. Introduction drift type 3 — mechanism question without mechanism test

Introduction:

Does cognitive load mediate the benefit?

Methods:

no cognitive-load measure.

82. Introduction drift type 4 — retrospective hypothesis

Unexpected subgroup result later becomes central Introduction rationale. This hides temporal order.

83. Introduction drift type 5 — population mismatch

Field problem framed around adolescents, study recruits adults.

Maybe legitimate—but the transition needs explanation.

84. Introduction–Methods handshake | Every promised noun needs a method

Introduction promise	Methods obligation
causal effect	causal design/identification
durability	delayed follow-up
mechanism	mechanism measure/manipulation
generalisation	sampling/context justification
comparison	defined comparator

85. Methods input contract | Methods receives the exact research question

Methods translates abstract concepts into observable procedures.

86. Methods output contract | It hands Results an analysis-ready evidence structure

By the end of Methods, the reader should know:

• who/what was studied;
• what conditions/exposures existed;
• what outcomes were measured;
• when they were measured;
• what data were included/excluded;
• how analyses answer the question.

87. Methods should make every primary Results sentence predictable in form | Not in value

Before seeing Results, the reader should know what kind of result will answer the primary question.

88. Methods drift type 1 — different outcome version

Introduction:

independent revision.

Methods:

same text revised with feedback still visible.

The operationalisation no longer matches “independent”.

89. Methods drift type 2 — intervention bundle hidden under one label

If X includes more time, more words, different teacher contact and different prompts, the Discussion cannot later attribute the effect to prompts alone.

90. Methods drift type 3 — population eligibility changes after recruitment

Post-hoc exclusion can alter the inference population.

91. Methods drift type 4 — analysis described too vaguely

Data were analysed statistically.

Results later report adjusted regression estimates, leaving no clear bridge.

92. Methods drift type 5 — preprocessing invisible

Computational work may transform raw data substantially. Hidden preprocessing creates an unreported link in the chain.

93. Methods drift type 6 — missing data plan omitted

Results later exclude 18% of cases with no explanation.

94. Methods drift type 7 — primary analysis not identified

Results contains ten models; reader cannot tell which was planned to answer the main question.

95. Methods–Results handshake | Every analysis should have a result; every result should have a method

This is one of the strongest paper-level audits.

96. Orphan Method | Method with no result

Possible reasons:

• result omitted;
• method no longer needed;
• space constraint;
• analysis failed;
• selective reporting.

Resolve transparently.

97. Orphan Result | Result with no Method parent

Often indicates:

• post-hoc analysis;
• undeclared preprocessing;
• missing outcome definition;
• hidden subgroup.

98. Results input contract | Results receives predefined evidence objects

These include samples, outcomes, models, themes or comparisons defined in Methods.

99. Results output contract | It hands Discussion a faithful evidence hierarchy

Results should clearly distinguish:

• participant/sample flow;
• primary finding;
• secondary findings;
• null/negative findings;
• exploratory findings;
• sensitivity/robustness findings.

100. Results should not create a new research question by emphasis | Emphasis can mutate the paper

If the primary result is null but the most visually prominent figure is a positive post-hoc subgroup, the paper’s apparent question has shifted.

101. Results drift type 1 — denominator inconsistency

Text n differs from table n with no explanation.

102. Results drift type 2 — outcome naming changes

Methods:

self-efficacy.

Results:

confidence.

If scales differ conceptually, terminology drift misleads.

103. Results drift type 3 — primary/secondary hierarchy lost

104. Results drift type 4 — significance replaces magnitude

Methods defines a meaningful difference; Results reports only p-values.

105. Results drift type 5 — adjusted vs unadjusted confusion

Table title and prose should make clear which estimate is primary.

106. Results drift type 6 — unexplained subgroup

Where did the subgroup definition come from?

107. Results drift type 7 — exploratory analysis written in confirmatory tone

108. Results drift type 8 — figure and text disagree

If figure suggests n = 80 and text claims n = 100, investigate before publication.

109. Results–Discussion handshake | Discussion cannot interpret a result that Results did not report

If Discussion spends a paragraph on a subgroup effect absent from Results, the reader cannot inspect the evidence.

110. Discussion input contract | Discussion receives the complete evidence hierarchy

Not only positive findings.

111. Discussion output contract | It hands Conclusion an evidence-tested interpretation

Discussion should produce:

• primary answer;
• interpretive meaning;
• relationship to prior evidence;
• rival explanations;
• limitations;
• scope;
• implications;
• remaining uncertainty.

112. Discussion drift type 1 — mechanism invented

Result:

performance improved.

Discussion:

metacognition increased.

No metacognition measure.

113. Discussion drift type 2 — causal language upgraded

Observational result becomes “X caused Y”.

114. Discussion drift type 3 — null result minimised

Primary null disappears behind secondary positive interpretation.

115. Discussion drift type 4 — limitation listed but not propagated

Limitation exists in paragraph 8, but conclusion strength remains unchanged.

116. Discussion drift type 5 — literature relation misclassified

A partial match becomes “replicates”. A conflicting result becomes “consistent”.

117. Discussion drift type 6 — sample becomes population

External validity expands without evidence.

118. Discussion drift type 7 — practical recommendation outruns decision evidence

119. Discussion–Conclusion handshake | Conclusion inherits the surviving claim

The Conclusion should not reopen rejected interpretations.

120. Conclusion input contract | Conclusion receives the final tested evidence state

It should receive a claim that already survived:

• counterevidence;
• limitations;
• alternative explanations;
• scope constraints;
• uncertainty.

121. Conclusion output contract | It gives the reader the paper’s final stable state

Answer + contribution + boundary + next implication.

122. Conclusion drift type 1 — certainty jump

123. Conclusion drift type 2 — scope jump

124. Conclusion drift type 3 — recommendation jump

125. Conclusion drift type 4 — new theory

126. Conclusion drift type 5 — new data

127. Conclusion drift type 6 — generic “importance” replaces contribution

128. Full relay audit | Ask one question at every handoff

Handoff	Audit question
Title → Abstract	Does Abstract deliver the title’s claim?
Abstract → Introduction	Does Introduction justify the same problem?
Introduction → Methods	Can Methods answer the stated question?
Methods → Results	Does every primary method produce a reported result?
Results → Discussion	Are interpretations anchored in reported results?
Discussion → Conclusion	Does Conclusion preserve the tested scope and certainty?

129. Cross-section coherence is strongest when each handoff is boring | No surprise identities

The findings themselves can be surprising. The identity of the study should not be.

130. Part II operating rule | 每一个 section 都要把同一个 study 传下去

A coherent manuscript is one in which every section changes the representation of the study without changing the identity of the study.

每个 section 可以换 representation，但不能换 research identity。

---

Part II traced the handoffs. Part III will turn coherence into a line-by-line audit system: claim lineage, number reconciliation, terminology control, figure/table consistency, protocol consistency and the detection of silent manuscript drift.

Part III — The manuscript audit system | 第三部分：把整篇论文变成可审计系统

131. Claim lineage | Every major claim needs a parent chain

For every major sentence in Abstract, Discussion or Conclusion, trace backwards:

claim → result → analysis → measure → construct → research question.

If one step is missing, the claim is not fully grounded.

132. Claim-lineage table | 一张表找出 orphan claims

Final claim	Result parent	Analysis parent	Measure parent	Question parent
near-term revision improved	one-week adjusted difference	baseline-adjusted regression	unseen revision rubric	does structured feedback improve one-week transfer?
mechanism = action clarity	no direct mechanism result	none	none	not asked

The second claim fails immediately.

133. Orphan-claim audit | Highlight sentences with no measurable ancestor

Common orphan claims:

• the intervention increased metacognition;
• the programme builds independence;
• the model is trustworthy;
• the policy improves equity;
• the method is sustainable.

These may be reasonable hypotheses or implications, but they need explicit evidence parents before being stated as findings.

134. Claim ancestry can be indirect but must be visible | Interpretation is allowed

A Discussion claim does not need to repeat the measurement label. It may interpret. The chain simply needs to show how the interpretation grows from evidence and where additional assumptions enter.

135. Assumption tagging | Mark every step that adds something not directly measured

Example:

revision score ↑ → independent revision capability ↑ → learner autonomy ↑.

The first arrow may be strong; the second may require a broader construct assumption.

136. Number reconciliation | All n values should have a story

Create a participant/sample ledger:

Stage	n	Reason for change
screened	142	starting pool
eligible	126	16 excluded by criteria
randomised	120	6 declined before allocation
primary outcome	112	8 missed follow-up
12-week outcome	97	15 additional attrition

137. Reconcile every number across Abstract, Methods, Results, figures and supplement

An abstract may correctly say “120 participants were randomised” while Results correctly analyse 112. The manuscript must make the difference legible.

138. Percentage reconciliation | Numerator and denominator travel together

Write:

18 of 24 participants (75%).

not simply:

75%.

when the denominator is small or changes across analyses.

139. Outcome ledger | One row per outcome and time point

Outcome	Priority	Time	Measure	Analysis	Reported?
independent revision	primary	1 week	rubric score	adjusted regression	yes
confidence	secondary	1 week	self-report scale	group comparison	yes
durability	secondary	12 weeks	rubric score	adjusted regression	yes

140. Outcome ledger detects switching | Protocol → manuscript comparison

If an outcome is primary in protocol but secondary in paper, explain why.

141. Time-point ledger | Same outcome at different times is not one result

Immediate, one-week and 12-week measures should remain distinguishable throughout.

142. Analysis ledger | One row per planned and exploratory model

Record:

• analysis question;
• planned/exploratory status;
• outcome;
• predictors;
• covariates;
• sample;
• reported location.

143. Analysis lineage reveals silent model switching

If Methods names Model A and Results foregrounds Model C, readers should know why.

144. Model numbering should mean the same thing everywhere

“Model 2” in Table 3 should not differ from “Model 2” in Supplementary Figure 1.

145. Covariate consistency | Adjustment set should be stable or explained

If age and baseline score appear in one model but motivation appears later, distinguish prespecified from exploratory adjustment.

146. Direction consistency | Positive must mean the same thing

Check scales where lower scores mean improvement. A coefficient sign can be interpreted backwards if direction is forgotten.

147. Unit reconciliation | Units should never rely on memory

Create a unit table for:

• time;
• distance;
• mass;
• concentration;
• currency;
• scores;
• rates;
• percentages.

148. Percentage vs percentage-point audit

A change from 20% to 30% is:

• 10 percentage points;
• 50% relative increase.

Do not switch between them silently.

149. Relative vs absolute effect audit | Especially important for decisions

Abstract and Conclusion should not foreground only the more dramatic relative framing when absolute effect is decision-relevant.

150. Precision audit | Decimal places must agree with measurement

If a rubric scores whole points, reporting 73.42819 can create false precision.

151. Rounding consistency | Same estimate should not look different

2.46 in Results and 2.5 in Abstract may be fine, but establish a rounding rule.

152. Confidence-interval consistency | Same estimate, same interval

Copying errors are common when manuscripts evolve. Reconcile tables and prose from one source of truth.

153. p-value consistency | Avoid manual transcription drift

If p-values are required, generate them from the analysis output rather than typing from memory across multiple sections.

154. Figure–text audit | What does the figure actually show?

For each figure:

1. state the question it answers;
2. state the sample displayed;
3. state the outcome/time point;
4. state the uncertainty display;
5. check that prose highlights the same pattern.

155. Figure-title audit | A title should not overinterpret

Weak:

Figure 2. Structured Feedback Produces Durable Learning.

If figure shows a four-week score:

Figure 2. Revision Scores by Condition Across the Four-Week Follow-up.

156. Axis audit | Axis label must match variable and unit

157. Legend audit | Error bars, groups and symbols must be defined

158. Colour-independent meaning | Visual accessibility and redundancy

Use labels, patterns or shapes where necessary so meaning is not carried by colour alone.

159. Table–text audit | Table gives detail; prose gives pattern

Do not repeat every cell. But do not claim a pattern that the table does not support.

160. Table denominator audit | n may differ by row

If missing data create different denominators, state them.

161. Table footnote audit | Important exceptions should not hide in tiny notes

If an analysis uses a different population or method, main text may need to mention it.

162. Supplement audit | Supplement is not a parallel universe

Check whether supplementary analyses:

• change the main conclusion;
• reveal missing outcomes;
• contradict primary text;
• contain key limitations.

163. Protocol consistency audit | Where applicable

Compare:

• primary outcome;
• secondary outcomes;
• time points;
• sample size;
• analysis plan;
• subgroups;
• stopping rules.

164. Registry consistency audit | Trial registry or preregistration

Differences are not automatically wrongdoing. Undeclared differences are the problem.

165. Protocol deviation statement | Name the change and timing

Useful pattern:

The protocol specified X. Before outcome analysis, we changed to Y because Z. This deviation is reported here and in the updated analysis plan.

166. Exploratory addition statement | Do not erase discovery

After inspecting the primary result, we conducted an exploratory analysis of baseline proficiency as a potential moderator.

167. Data-source consistency | Same version of the data?

Lock or version datasets used for final analysis.

168. Code–paper consistency | Computational manuscripts

Code variable names may differ from prose, but maintain a mapping.

169. Model-version consistency | AI/ML systems

If model providers update systems, record the exact version/date where possible.

170. Prompt/procedure consistency | AI-assisted research

The prompt shown in Supplement should match the procedure described in Methods.

171. Dataset-split consistency | Train/validation/test identities must not blur

Performance on validation data should not be reported as test performance.

172. Benchmark consistency | Same metric definition across comparisons

Accuracy, macro-F1 and weighted-F1 answer different questions.

173. Qualitative codebook consistency | Theme names should stabilise

Methods:

action clarity.

Results:

directional guidance.

Discussion:

cognitive scaffolding.

These may represent different analytical levels. Explain transitions.

174. Quote attribution consistency | Speaker labels and participant IDs

Do not accidentally change participant identity across text and tables.

175. Mixed-methods integration consistency | Strands must refer to comparable phenomena

If quantitative outcome measures performance and qualitative interviews discuss satisfaction, do not call them “convergent evidence for learning” without a bridge.

176. Systematic-review study-count consistency | Records vs reports vs studies

These are different units. PRISMA-style flow makes the distinction explicit.

177. Meta-analysis effect-measure consistency | Standardised mean difference vs raw mean difference

Do not interpret them interchangeably.

178. Risk-of-bias consistency | Study quality should influence Discussion certainty

If most included studies are high risk of bias, Conclusion should not sound like high-certainty evidence.

179. Citation consistency | Does the cited source support the exact claim?

Cross-section coherence also includes source claims. A citation in Introduction may define a result one way while Discussion characterises it differently.

180. Citation-distance audit | Primary vs secondary source

For central empirical claims, check whether you are relying on a review’s summary when the primary study is needed.

181. Terminology dictionary | Create a manuscript lexicon

Preferred term	Definition	Avoid unless distinct
independent revision	revision of unseen task without feedback	writing ability, autonomy
structured feedback	three action-oriented prompts	enhanced support
baseline proficiency	pre-study proficiency assessment	prior knowledge, ability

182. Bilingual terminology dictionary | Chinese bridge + English technical term

For Mandarin-speaking teams, maintain a paired lexicon so Chinese notes do not introduce new English constructs during drafting.

183. Chinese synonym drift | 中文内部先漂移，英文再放大

Example:

学习效果 / 学习能力 / 学业表现 / 掌握程度

These are not automatically the same construct. Decide before translating.

184. “效果” is especially dangerous | Effect, performance, outcome, benefit?

English requires a more specific relation.

185. “能力” is broad | Capability may exceed measured task performance

Use the measured construct unless transfer evidence supports capability language.

186. “提升” can hide comparison and time | Improvement relative to what?

Baseline? Control? Previous version? Expected trend?

187. “有效” can hide outcome | Effective for what?

188. “显著” can hide statistical vs practical meaning

189. “稳定” can hide time/sample/model dimension

190. “长期” needs an actual horizon

191. The bilingual invariant card | Lock concepts before English drafting

For each core term record:

• Chinese working label;
• English technical label;
• operational definition;
• allowed near-synonyms;
• forbidden broader terms.

192. Revision-history audit | Manuscript version control

Major revisions can create internal contradictions. When one section changes, identify dependent sections.

193. Change-propagation rule | A change has downstream owners

If primary outcome changes:

• title may change;
• abstract must change;
• Introduction aim may change;
• Methods outcome hierarchy must change;
• Results ordering must change;
• Discussion emphasis must change;
• Conclusion claim must change.

194. Change map | Treat manuscript edits as a dependency graph

This is why “small” edits can create large inconsistencies.

195. Find–replace is not enough | Conceptual change needs semantic propagation

Changing “confidence” to “self-efficacy” everywhere may still be wrong if the scale does not measure self-efficacy.

196. Cross-reference audit | Section/Table/Figure references

Check that references point to the correct item after revisions.

197. Appendix/supplement reference audit

198. Reference-list audit | Every citation has an entry; every entry is cited where required

199. Reporting-guideline audit | Use the design-appropriate checklist

CONSORT 2025, STROBE, PRISMA, SRQR/COREQ, TRIPOD and other guidelines help expose missing reporting obligations. They are not substitutes for design quality, but they can identify manuscript-level omissions.

200. Journal-format audit | Genre structure can differ

Nature Methods, for example, structures Articles around Introduction, Results, Discussion and Online Methods, while other venues use classic IMRaD or combined Results/Discussion. Coherence is functional rather than dependent on identical headings.

Nature Methods | Content Types and Article Structure

201. Accessibility audit | Expert accuracy + reader navigation

Nature Computational Science emphasises making core manuscript sections accessible while preserving technical information. Coherence helps accessibility because readers do not have to relearn the study’s terms and identities in every section.

Nature Computational Science | On Writing Accessible Computational Science Papers

202. The “same noun” audit | Search core nouns across manuscript

Search:

• population label;
• intervention name;
• primary outcome;
• time points;
• main construct;
• design name.

Inspect every variant.

203. The “same number” audit | Search every major n and estimate

204. The “same verb” audit | Search causal verbs

Search:

• causes;
• improves;
• leads to;
• predicts;
• associated with;
• suggests;
• supports.

Check whether section-to-section certainty is stable.

205. The “same time” audit | Search week/day/month labels

206. The “same outcome” audit | Search primary outcome name and all synonyms

207. The “same sample” audit | Search n and inclusion labels

208. The “same analysis” audit | Search model/test names

209. The “same scope” audit | Search broad nouns in Discussion/Conclusion

Look for:

• learners;
• students;
• patients;
• users;
• schools;
• companies;
• systems.

Compare with actual sample.

210. Part III operating rule | Coherence can be audited mechanically and conceptually

The manuscript should have one source of truth for numbers, one stable identity for constructs, one explicit hierarchy for outcomes and one traceable lineage for every major claim.

整篇 paper 最好有：numbers 的 single source of truth、construct 的 stable identity、outcomes 的 explicit hierarchy、以及每一个 major claim 的 traceable lineage。

---

Part III built the audit system. Part IV will run that system on a full fictional manuscript, showing how a paper can drift across sections and how to repair it without rewriting everything from scratch.

Part IV — Full manuscript repair: from a drifting paper to one coherent study | 第四部分：完整 paper 修复

All study details in this worked example are fictional teaching material.

211. The fictional study | Study identity before drift

Research problem:

Advanced bilingual learners often revise successfully when feedback is visible, but it is unclear whether structured action-oriented feedback transfers to an unseen task after support is removed.

Design:

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

Primary outcome:

one-week independent revision score on an unseen argumentative task.

Secondary outcomes:

self-reported confidence, evidence-integration subscore, sentence accuracy and a 12-week revision score.

Primary analysis:

baseline-adjusted linear regression comparing groups on the one-week score.

212. The intended paper identity card

Field	Identity
Population	advanced bilingual adult learners
Design	parallel randomised trial
Intervention	structured action-oriented feedback
Comparator	standard evaluative feedback
Primary outcome	one-week unseen revision score
Secondary outcomes	confidence, subscales, 12-week score
Main claim scope	near-term independent revision

213. Now deliberately break the manuscript | 漂移版本

The following sections are individually plausible but collectively inconsistent.

214. Broken title

Structured Feedback Builds Lasting Independent Writers

Problems:

• “builds” makes a strong causal and developmental claim;
• “lasting” exceeds the primary one-week endpoint;
• “independent writers” exceeds the measured revision construct.

215. Broken abstract

Structured feedback improved learner independence and confidence. Participants receiving structured feedback performed better and reported greater self-efficacy, demonstrating that the intervention develops durable writing ability. These findings support broad adoption of structured feedback in advanced English programmes.

Problems:

• primary outcome not named;
• confidence/self-efficacy terminology drift;
• durability asserted without strong delayed evidence;
• writing ability broader than revision score;
• adoption recommendation not supported by routine-setting evidence.

216. Broken Introduction

Feedback is essential for autonomous learning, yet little research has examined whether structured feedback develops independent writers. We therefore tested whether structured feedback improves long-term writing ability and learner confidence.

Problems:

• research question now includes long-term writing ability;
• confidence is elevated to co-primary status without protocol basis;
• “autonomous learning” introduced as broad construct;
• gap no longer matches actual primary outcome.

217. Broken Methods

Participants received either structured or standard feedback. Writing outcomes were assessed after the intervention. Data were analysed using regression.

Problems:

• intervention not reconstructable;
• one-week primary timing hidden;
• outcome hierarchy hidden;
• 12-week timing not distinguished;
• primary model and adjustment not stated;
• independent unseen-task definition absent.

218. Broken Results

Confidence was significantly higher in the structured group (p = .01). The structured group also performed better on revision, especially among lower-proficiency learners. These results demonstrate the effectiveness of structured feedback.

Problems:

• secondary confidence result appears first;
• primary estimate/magnitude/uncertainty absent;
• post-hoc subgroup promoted;
• causal interpretation appears in Results;
• 12-week result omitted.

219. Broken Discussion

The findings show that structured feedback develops learner autonomy because clear action prompts increase metacognitive awareness. This confirms theories of scaffolding and demonstrates that advanced bilingual learners need structured guidance to become independent writers. Although the study was conducted in one programme, the implications are likely to apply broadly to advanced English education.

Problems:

• autonomy/metacognition not directly measured;
• mechanism invented;
• “confirms” overstates theory discrimination;
• population/generalisation inflated;
• delayed uncertainty absent;
• single-site limitation stated but erased.

220. Broken Conclusion

In conclusion, structured feedback is an effective and durable approach that develops independent writers. English programmes should therefore adopt structured feedback widely.

Problems:

• certainty jump;
• duration jump;
• construct jump;
• scope jump;
• recommendation jump.

221. Why this broken paper can still look “academic” | Fluency hides drift

Every section contains plausible academic vocabulary: autonomous learning, significant, demonstrates, metacognitive awareness, implications. The paper fails not because the language is simple but because the research identity changes repeatedly.

222. Repair strategy 1 — Freeze the identity card

Before rewriting prose, lock:

• population;
• design;
• intervention;
• comparator;
• primary outcome;
• time point;
• primary analysis;
• exploratory status of subgroup;
• 12-week uncertainty.

223. Repair strategy 2 — Restore evidence hierarchy

Primary one-week revision first.

Confidence secondary.

Subgroup exploratory.

12-week result retained because it changes durability interpretation.

224. Repair strategy 3 — Separate constructs

Independent revision score is measured.

writing ability is broader.

learner autonomy is broader still.

metacognition is a possible mechanism, not measured.

225. Repair strategy 4 — Restore temporal labels

Immediate supported task.

One-week independent task.

12-week follow-up.

Each must remain separate.

226. Repair strategy 5 — Restore causal boundaries

Randomisation supports a causal effect of assignment to the full structured condition on the one-week measured outcome. It does not isolate which intervention component caused the effect.

227. Repair the title | Version 2

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

This title:

• keeps population;
• keeps outcome;
• keeps time;
• signals design;
• does not claim mechanism or durability.

228. Repair the Abstract | Version 2

We tested whether structured action-oriented feedback improves independent revision after feedback removal in advanced bilingual adult learners. In a randomised two-group trial, 120 learners received structured or standard evaluative feedback; the primary outcome was revision quality on an unseen task one week later. Structured feedback produced a higher one-week revision score, with the largest difference in evidence integration, while the 12-week estimate was smaller and imprecise. The findings support a near-term independent-revision benefit under the studied conditions but do not establish long-term durability or the specific active mechanism.

229. Abstract repair checklist

Question restored.

Design restored.

Primary outcome restored.

Delayed uncertainty restored.

Mechanism boundary restored.

230. Repair the Introduction | Version 2

Structured feedback can improve revision while learners work directly from feedback, but supported performance does not establish that revision decisions transfer after support is removed. Existing work therefore leaves a narrower question unresolved: whether structured action-oriented feedback improves performance on an unseen independent task after a short delay. The present study tested the one-week independent-revision effect in advanced bilingual adult learners, with delayed performance and confidence examined as secondary outcomes.

231. Introduction repair lesson | The study becomes smaller and clearer

The gap is now exactly the gap the design can address.

232. Repair the Methods | Version 2

We conducted a parallel randomised trial with 120 advanced bilingual adult learners. Participants were allocated 1:1 to structured action-oriented feedback or standard evaluative feedback. The structured condition organised feedback into claim, evidence and reasoning actions; the comparator identified strengths and weaknesses without required action prompts. The primary outcome was revision quality on an unseen argumentative task completed one week after feedback removal and scored by blinded raters using a preregistered rubric. The primary analysis estimated the between-group difference using linear regression adjusted for baseline revision score. Confidence, rubric subdimensions and a 12-week unseen revision score were secondary outcomes. Baseline-proficiency subgroup analyses were exploratory.

233. Methods repair lesson | Reader can now predict Results structure

The reader expects:

1. sample flow;
2. one-week adjusted group difference;
3. secondary confidence/subdimensions;
4. 12-week result;
5. exploratory subgroup.

234. Repair the Results | Version 2

Of 120 randomised learners, 112 completed the one-week primary outcome. The structured-feedback group scored a mean of 74.1 compared with 68.2 in the standard-feedback group; the baseline-adjusted mean difference was 5.7 points (95% CI 2.3–9.1). The largest rubric difference occurred in evidence integration, while sentence-level accuracy differed little. Confidence was higher in the structured group as a secondary outcome. At 12 weeks, the adjusted difference was 1.3 points (95% CI −2.0 to 4.6). In an exploratory post-hoc analysis, the one-week difference was larger among lower-baseline learners, although the interaction estimate was imprecise.

235. Results repair lesson | Every line has a Method parent

No new construct appears. Hierarchy is visible. Null/imprecise delayed evidence remains present.

236. Repair the Discussion | Version 2

The primary finding supports a near-term independent-revision benefit from the structured feedback condition, extending earlier evidence from supported revision tasks. The concentration of the difference in evidence integration suggests that higher-order revision decisions were more affected than sentence-level accuracy. Participant confidence also increased, but confidence was a secondary self-report outcome and should not be treated as evidence of broader writing competence. The structured condition required more explicit planning, so the present design does not isolate action structure as the unique mechanism. The smaller and imprecise 12-week estimate leaves durability unresolved. Because the trial was conducted in one high-support programme with advanced adult volunteers, transportability to younger, beginner or lower-support settings remains uncertain.

237. Discussion repair lesson | Interpretation grows from measured evidence

The paper can discuss a plausible higher-order effect without inventing autonomy or metacognition as established mechanisms.

238. Repair the Conclusion | Version 2

Structured action-oriented feedback improved one-week independent revision relative to standard evaluative feedback in this advanced bilingual sample. The study contributes delayed unaided evidence beyond supported-task improvement, with the clearest benefit in evidence integration. The active mechanism and longer-term durability remain unresolved, and the single high-support setting limits broader transportability. Component-controlled and multi-site delayed replication are therefore the next critical tests before routine implementation claims are made.

239. Full-paper coherence after repair | The identity survives

Section	Primary identity preserved?	Main boundary preserved?
Title	yes	one week + population
Abstract	yes	durability/mechanism
Introduction	yes	near-term transfer question
Methods	yes	operational definitions
Results	yes	hierarchy + uncertainty
Discussion	yes	mechanism/scope
Conclusion	yes	durability/transportability

240. Repair principle: change the smallest upstream point that fixes the downstream chain

If Discussion overclaims because the Introduction question was already too broad, fixing only Discussion wording may not be enough. Repair the earliest point where the identity drift begins.

241. Upstream repair map

Observed problem	Likely earliest repair point
Conclusion too broad	Discussion scope or Introduction framing
Result does not answer question	Methods operationalisation or Introduction question
Mechanism unsupported	Methods lacked mechanism test
Outcome switching	protocol/Introduction hierarchy
n mismatch	sample-flow ledger

242. Do not “repair” by deleting inconvenient evidence | Coherence is not narrative cleaning

If the 12-week result weakens durability, keep it. The correct repair is to narrow the claim, not hide the result.

243. Do not “repair” by renaming the construct everywhere | Semantic repair, not cosmetic consistency

If confidence is not competence, changing every occurrence of “confidence” to “competence” makes the paper consistently wrong.

244. Do not “repair” by adding disclaimers only at the end | Boundaries should propagate

A single “limitations” sentence cannot compensate for causal language in title, abstract and Discussion.

245. Coherence repair is dependency repair | One change propagates

When the 12-week effect becomes uncertain, update:

• Abstract conclusion;
• Discussion durability claim;
• Conclusion;
• possibly title if it says “lasting”.

246. Version-control habit | Write a change note for major conceptual edits

Example:

Changed main scope from durable writing ability to near-term independent revision after delayed result review. Updated title, abstract, Introduction aim, Discussion and Conclusion.

247. Team-writing risk | Multiple authors create local coherence and global drift

One author writes Introduction, another Results, another Discussion. Each may use different construct names and assumptions.

248. Team solution: shared identity card + lexicon + outcome ledger

Do not rely on final proofreader to infer scientific identity from prose.

249. Senior-author synthesis pass | Read horizontally, not vertically

Instead of reading page 1 to page 20 only, read:

1. all statements of research question;
2. all primary-outcome mentions;
3. all causal claims;
4. all population labels;
5. all contribution claims.

This horizontal read exposes drift quickly.

250. Reviewer-style horizontal read | “Same claim everywhere?”

Copy the title, Abstract conclusion, final Introduction aim, primary Results sentence, Discussion opening and Conclusion into one page. Compare them side by side.

251. The six-surface coherence sheet

Surface	One-sentence study statement
Title	what the paper appears to claim
Abstract	what the paper claims to show
Introduction	what it intends to test
Results	what it actually found
Discussion	what it thinks the findings mean
Conclusion	what survives as final claim

252. If these six sentences describe six different studies, stop editing sentences | Repair research identity first

253. Bilingual full-paper drift | Chinese notes can create cross-section mismatch

A team may use Chinese shorthand internally:

效果不错 / 提升明显 / 长期有帮助 / 学习能力增强

Different writers may reconstruct each phrase differently in English, producing inconsistent claims.

254. Bilingual team solution | Translate the identity card, not free-form shorthand

Use exact paired labels:

one-week independent revision score = 一周后无反馈独立修改得分

Keep the construct precise in both languages.

255. The “Chinese note → English manuscript” risk map

Chinese shorthand	Possible English drift
学习效果	learning outcome / performance / effectiveness
能力	ability / capability / competence
显著	statistically significant / substantial / clear
有效	effective / efficacious / useful
稳定	robust / persistent / consistent

256. Repair by operational definition, not by preferred English sophistication

257. Full-manuscript claim ladder | Final horizontal map

Data: 5.7-point adjusted difference.

Result: one-week revision higher.

Study inference: structured condition improved measured one-week revision.

Construct inference: near-term independent revision benefited.

Mechanism: unresolved.

Durability: unresolved.

Implementation: untested broadly.

258. Every section should stop at the right rung | Not every section climbs equally high

Results stays low.

Discussion may climb higher with caveats.

Conclusion compresses the highest rung that survived.

259. The final repaired paper is not “weaker” | It is more citable

Readers can now know exactly what the study supports. Precision increases usefulness.

260. Part IV operating rule | Repair identity before style

When a manuscript feels inconsistent, do not start by polishing transitions. Reconstruct the study identity, restore evidence hierarchy and repair the earliest broken handoff. Style comes after the research objects agree.

如果整篇论文“哪里不对”，先不要润色 transitions。先重建 study identity、evidence hierarchy，并修复最早断掉的 handoff。研究对象一致以后，style 才有意义。

---

Part IV repaired one full manuscript. Part V will broaden the system across qualitative, mixed-methods, systematic-review, computational and professional reports, where “coherence” has different visible forms but the same underlying logic.

Part V — Coherence across research genres | 第五部分：不同研究类型怎样保持同一个 paper identity

261. Coherence is functional, not tied to one heading pattern

Some journals use IMRaD. Others place Methods online. Some combine Results and Discussion. Qualitative work may integrate findings and interpretation. Computational papers may foreground benchmark architecture. Coherence therefore means stable research identity across functions, not identical visible section labels.

262. Qualitative paper identity | Question → sampling → data generation → analysis → theme → interpretation

A qualitative manuscript still needs lineage. If the Introduction asks how postgraduate writers experience supervisory feedback, Methods should recruit participants relevant to that experience, Results should present themes grounded in those accounts, and Discussion should interpret those themes without turning them into population prevalence or causal effects.

263. Qualitative drift type 1 — research question changes from meaning to frequency

Introduction asks:

How do writers experience feedback overload?

Results concludes:

70% of students prefer concise feedback.

The study may not have been designed for population frequency inference.

264. Qualitative drift type 2 — theme becomes mechanism

Theme:

participants describe clearer next actions.

Discussion:

action clarity causes better writing.

The qualitative evidence may support a conceptual explanation, not a causal effect.

265. Qualitative drift type 3 — sample diversity disappears in Conclusion

Purposive sample of one programme becomes “multilingual students”.

266. Qualitative coherence card

Invariant	Qualitative form
Question	experience/process/meaning
Sample	sampling logic + context
Data	interviews/observations/texts
Analysis	coding/thematic/interpretive method
Result	themes/patterns/deviant cases
Conclusion	conceptual pattern + transferability boundary

267. Mixed-methods paper identity | Integration must be planned and reported

A mixed-methods paper contains at least two evidence strands, but it still needs one coherent research architecture.

268. Mixed-methods question architecture

Possible structure:

• quantitative question: did performance change?
• qualitative question: how did participants experience the intervention?
• integration question: how do the two strands jointly explain implementation and outcome?

269. Mixed-methods drift type 1 — strands answer unrelated questions

Quant measures writing performance; interviews ask about course enjoyment; Discussion calls them “convergent evidence for learning”. The bridge is missing.

270. Mixed-methods drift type 2 — one strand becomes decorative

Methods promises equal integration, but Discussion uses interviews only as quotations supporting the quantitative story.

271. Mixed-methods drift type 3 — contradiction is hidden

Scores improve, but participants report overload. Coherence requires representing both rather than forcing convergence.

272. Mixed-methods integration ledger

Question	Quant evidence	Qual evidence	Integrated inference
Does performance improve?	revision score	not primary	quant answers directly
How is structure experienced?	usability scale	interviews	clarity + overload trade-off
Why might outcomes differ?	subscale pattern	process accounts	hypothesis, not proved mechanism

273. Mixed-methods conclusion must preserve evidence type

Do not turn participant explanation into causal proof just because it aligns with quantitative outcomes.

274. Systematic-review identity | Question → eligibility → search → selection → appraisal → synthesis → conclusion

Systematic reviews have a distinctive cross-section contract. The Introduction question must match eligibility criteria; Methods search and selection must identify the studies; Results must report the evidence actually found; Discussion certainty must reflect heterogeneity and risk of bias.

275. Review drift type 1 — eligibility criteria narrower than Introduction claim

Introduction asks about “all digital learning interventions”; Methods includes only English-language randomised school studies.

Conclusion should not return to “all digital learning”.

276. Review drift type 2 — records vs studies confused

5,000 search records are not 5,000 studies.

277. Review drift type 3 — pooled effect erases heterogeneity

One average number should not become a universal effect if study populations and outcomes differ strongly.

278. Review drift type 4 — risk of bias disappears in Conclusion

If most evidence is low certainty, final wording should reflect that.

279. PRISMA-style flow as coherence device | Search identity becomes visible

Flow diagrams help reconcile records identified, screened, excluded and included. The visual is part of manuscript identity.

280. Meta-analysis coherence | Effect measure must remain stable

If some outcomes use odds ratios and others use standardised mean differences, synthesis and interpretation should preserve which metric supports which conclusion.

281. AI/ML paper identity | Task → dataset → model → split → metric → benchmark → interpretation

Computational papers are especially vulnerable to identity drift because many versions, datasets and metrics coexist.

282. AI drift type 1 — validation performance becomes test performance

If hyperparameters were selected using validation data, that performance is not an independent test estimate.

283. AI drift type 2 — benchmark performance becomes general intelligence or utility

A benchmark-specific gain should remain benchmark-specific unless broader validation exists.

284. AI drift type 3 — model version changes mid-study

If a hosted model changes between data collection and replication, document the version/time and avoid treating outputs as one stable system.

285. AI drift type 4 — prompt changes without manuscript update

The prompt/procedure shown in Methods must match the one used to generate study outputs.

286. AI drift type 5 — metric cherry-picking

Introduction promises reliability, Results foreground accuracy, Discussion concludes trustworthiness. Reliability, accuracy and trustworthiness are not identical.

287. AI coherence card

Invariant	Computational form
Task	prediction/classification/generation objective
Data	dataset version + split
Model	architecture/version/settings
Metric	exact definition
Benchmark	evaluation context
Claim	bounded to tested conditions

288. TRIPOD-LLM as a cross-section reporting map

TRIPOD-LLM’s section-spanning items illustrate why LLM studies require title/abstract clarity, detailed Methods, transparent Results and appropriately bounded Discussion rather than isolated disclosure. The manuscript must tell one stable story about the model, task, data and evaluation.

289. Engineering paper identity | Requirement → design → operating condition → metric → trade-off → deployment boundary

Engineering coherence often depends on operating envelope.

290. Engineering drift type 1 — lab condition becomes field claim

Tested at 20–30°C.

Conclusion:

robust in industrial environments.

Operating range exceeds evidence.

291. Engineering drift type 2 — one metric becomes “better”

Throughput rises while energy use and error rate worsen. Conclusion must represent multi-objective trade-offs.

292. Engineering drift type 3 — prototype becomes scalable system

Feasibility is not deployment readiness.

293. Diagnostic-study identity | Index test → reference standard → population → threshold → sensitivity/specificity → use case

A diagnostic conclusion must remain tied to the population and threshold tested.

294. Diagnostic drift type 1 — high sensitivity becomes “accurate”

Specificity may be modest.

295. Diagnostic drift type 2 — prevalence changes ignored

Predictive values can change across populations with different prevalence.

296. Feasibility-study identity | Can the study/programme be done?

Recruitment, retention, adherence and delivery may be primary. Effectiveness estimates are often exploratory.

297. Feasibility drift | Pilot outcome becomes efficacy claim

A small feasibility study should not conclude that the intervention works because outcome means moved in the hoped-for direction.

298. Professional evaluation identity | Goal → intervention → implementation → outcome → cost/burden → decision

Professional reports may not use academic IMRaD, but they still need coherent decision logic.

299. Programme-evaluation drift type 1 — satisfaction becomes effectiveness

Parents like the programme; learner outcomes unchanged.

Conclusion:

successful.

Success criteria must be explicit.

300. Programme-evaluation drift type 2 — completion rate hides workload

If gains require unsustainable staff time, decision conclusions should represent the burden.

301. Policy-pilot identity | Policy objective → implementation → distributional effects → costs → risks → scale decision

Average benefit alone may be insufficient.

302. Policy drift type 1 — pilot becomes scale-up recommendation

Implementation conditions change with scale.

303. Policy drift type 2 — average effect hides subgroup harm

Distributional evidence must reach Discussion and Conclusion when decision-relevant.

304. Humanities/interpretive paper identity | Question → corpus/archive → interpretive method → textual evidence → claim → scope

Interpretive papers do not need to imitate experimental structure, but their claims still need stable evidence objects and scope boundaries.

305. Interpretive drift type 1 — selected corpus becomes genre-wide claim

306. Interpretive drift type 2 — motif becomes historical cause

Textual pattern can support interpretation without establishing causal social history.

307. Corpus-linguistics identity | Corpus definition → search/query → frequency/association → functional interpretation

308. Corpus drift type 1 — frequency becomes quality

Experts use phrase X more often, therefore phrase X is better English. This conclusion requires functional evidence beyond frequency.

309. Corpus drift type 2 — corpus composition ignored

If expert corpus contains different genres, frequency differences may reflect genre rather than expertise.

310. Cross-genre invariant | The evidence object must remain stable

Whether the evidence object is a regression coefficient, interview theme, pooled effect, benchmark score, engineering measurement or textual motif, the paper must keep its identity and scope visible.

311. Cross-genre invariant | The question–evidence fit must remain stable

A study designed to describe should not conclude causally. A study designed for feasibility should not conclude effectiveness. A study designed for interpretation should not claim prevalence without sampling logic.

312. Cross-genre invariant | Uncertainty must propagate

Low certainty in evidence must survive into Discussion and Conclusion regardless of genre.

313. Cross-genre invariant | Contribution must match the gap

Review gap, method gap, interpretation gap, implementation gap and benchmark gap are different contributions.

314. Cross-genre invariant | Section labels can change; claim lineage cannot

315. Bilingual genre drift | “研究结果” can mean different evidence types

In Chinese working notes, 研究结果 may refer to:

• quantitative estimate;
• qualitative theme;
• review synthesis;
• model performance;
• evaluation finding.

English reconstruction should preserve the evidence type.

316. Bilingual genre drift | “方法有效” can mean efficacy, usability or feasibility

Ask which evidence supports which claim.

317. Bilingual genre drift | “验证” is overloaded

验证 can mean:

• validate a model;
• verify a result;
• replicate;
• confirm a hypothesis;
• test a mechanism.

English must select the exact action.

318. Bilingual genre drift | “模型” can be statistical, conceptual or machine-learning model

Do not let one Chinese noun hide different technical objects.

319. Bilingual genre drift | “指标” can be metric, indicator, measure or outcome

320. Bilingual genre drift | “样本” can be specimen, sample or dataset

321. Build a genre-specific bilingual lexicon | Before drafting

For each technical object record:

• Chinese working term;
• English term;
• discipline-specific meaning;
• section where defined;
• allowed shorthand.

322. Journal adaptation without identity loss

You may shorten Methods, merge sections or move details to supplement to meet journal style. The study identity should remain intact.

323. Word-limit pressure creates coherence risk

When cutting:

• do not remove the primary-outcome definition;
• do not remove essential sample flow;
• do not remove the main uncertainty;
• do not remove the limitation that changes the conclusion.

324. Short-format paper | Compression priorities

Protect:

1. question;
2. design;
3. primary outcome;
4. main estimate;
5. final boundary.

325. Long thesis/dissertation | Expansion priorities

Expand:

• theory;
• method rationale;
• robustness;
• alternative explanations;
• limitations;

without changing the core study identity.

326. Multi-paper thesis risk | Each chapter has identity; thesis has meta-identity

A dissertation with several studies must distinguish chapter-level questions from the overarching research programme.

327. Registered-report coherence | Strong temporal trace

Registered Reports make the separation between planned questions/analyses and later findings especially explicit. The final manuscript can include exploratory work, but planned and exploratory identities remain distinct.

328. Preprint-to-journal coherence | Revision after peer review

If analyses change between versions, ensure title, abstract and conclusions update accordingly. Version history may matter for readers comparing preprint and final publication.

329. Correction coherence | Correct one number, propagate downstream interpretation

A correction to an estimate may require changes in Abstract, Discussion or Conclusion if meaning changes.

330. Retraction/correction distinction | Paper identity can change after publication

Post-publication notices should be interpreted according to what they actually correct or invalidate. A correction does not automatically invalidate the whole study; a retraction does not automatically prove the opposite claim.

331. Peer-review coherence | Reviewer asks should map to identity or evidence

When reviewers request new analyses, label them as post hoc if they were not prespecified. Do not rewrite Introduction to imply they were planned from the beginning.

332. Editor-driven title changes | Recheck scope

A shorter, punchier title can accidentally become broader or more causal.

333. Press-release drift | Outside-manuscript coherence matters too

Public summaries should not claim more than the paper. Although this lesson focuses on the manuscript, title, abstract and public-facing summaries all share the same evidence budget.

334. Graphical abstract drift | Visual simplification can distort

A single arrow X → Y may imply causation even if the study is observational. Visual language carries inferential force.

335. Lay-summary drift | Simpler language should not mean stronger claims

“Associated with” can be explained plainly without changing to “causes”.

336. The cross-genre paper identity card | Minimal universal fields

Field	Question
Research job	What kind of knowledge is this study trying to produce?
Evidence object	What counts as data/evidence here?
Population/corpus/context	What is directly represented?
Method	How was evidence produced/analysed?
Primary claim	What can be concluded?
Boundary	Where does that claim stop?

337. The universal coherence test | Same study or same-sounding prose?

Ask:

If I removed the section headings and read only the research objects—population, measure, time, analysis, result, claim—would they still belong to one study?

338. Genre-aware coherence is stronger than rigid IMRaD obedience

Do not force a qualitative or interpretive study into experimental conventions merely to look scientific. Preserve the correct evidence logic for the genre.

339. Section freedom increases the need for explicit handoffs | Combined sections need signalling

If Results and Discussion are combined, use clear move transitions so readers know when evidence reporting becomes interpretation.

340. Part V operating rule | 形式可以不同，identity 不能漂

Across genres, coherence means that the kind of question, kind of evidence, kind of analysis and kind of claim remain mutually compatible from the first page to the last.

不同研究类型可以有不同 section 结构，但 question、evidence、analysis 与 claim 的类型必须从第一页到最后一页彼此兼容。

---

Part V generalised coherence across genres. Part VI will turn the entire system into timed drills, team workflows, a 100-point manuscript-coherence rubric and a final unseen-paper benchmark.

Part VI — Team workflow, drills, mastery rubric and final independent benchmark | 第六部分：把整篇 paper coherence 变成可重复流程

341. Coherence should be designed before prose | 最好不要等到最后才找矛盾

The cheapest coherence error is the one prevented before drafting. Once seven sections, eight figures, three supplements and several co-authors exist, a small identity change becomes expensive to propagate.

342. Pre-draft workflow: create the study identity card

Before anyone writes prose, agree on:

• one-sentence research question;
• target population/context;
• design;
• intervention/exposure and comparator;
• primary outcome and time point;
• secondary outcomes;
• primary analysis;
• main causal/generalisation scope;
• planned contribution.

343. Pre-draft workflow: create the manuscript lexicon

List every central technical term and the exact meaning it carries.

344. Pre-draft workflow: create the outcome ledger

One row per outcome/time point. Mark primary, secondary and exploratory.

345. Pre-draft workflow: create the analysis ledger

One row per analysis. Mark planned/exploratory and the exact question each model answers.

346. Pre-draft workflow: create the figure/table map

Each display should have a job:

• participant flow;
• primary outcome;
• secondary pattern;
• robustness;
• mechanism/process;
• supplementary detail.

347. Drafting order does not have to equal reading order

Many researchers draft Methods and Results first, then Introduction and Discussion. That is fine if the final paper restores the reading-order logic and section handoffs.

348. One practical drafting sequence

1. Identity card.
2. Methods.
3. Results.
4. Figures/tables.
5. Introduction.
6. Discussion.
7. Conclusion.
8. Abstract.
9. Title.

This sequence allows title and abstract to inherit the completed evidence state.

349. Team assignment by section requires cross-section owners

Do not assign only “Introduction author” and “Methods author”. Also assign:

• study identity owner;
• numbers owner;
• terminology owner;
• analysis/result owner;
• final coherence reviewer.

350. Single source of truth for numbers | Prevent copy drift

Where possible, generate tables, figures and manuscript statistics from the same analysis outputs rather than hand-copying numbers into many places.

351. Single source of truth for terminology | Shared glossary

Especially useful for interdisciplinary and bilingual teams.

352. Single source of truth for outcome hierarchy | Protocol/analysis plan/identity card

353. The 10-minute coherence scan | Fast pre-submission check

1. Read title.
2. Read abstract conclusion.
3. Read final Introduction paragraph.
4. Read primary Methods outcome/analysis.
5. Read first primary Results paragraph.
6. Read Discussion opening.
7. Read Conclusion.

If these seven surfaces disagree, stop and repair.

354. The 20-minute horizontal audit

Time	Audit
3 min	question/aim mentions
3 min	population/sample labels
4 min	primary outcome + time point
4 min	causal verbs
3 min	n/denominators
3 min	contribution/scope claims

355. The 45-minute manuscript audit

Time	Task
8 min	build six-surface coherence sheet
8 min	reconcile sample flow and denominators
8 min	outcome/analysis ledger
7 min	figure/table prose reconciliation
7 min	terminology + bilingual lexicon
7 min	Discussion/Conclusion scope audit

356. The 90-minute deep audit

Time	Task
10 min	identity card reconstruction
10 min	protocol/registry comparison
15 min	claim-lineage map for all major claims
10 min	sample/outcome/analysis ledgers
10 min	figure/table/supplement reconciliation
10 min	causality + scope + construct audit
10 min	bilingual terminology audit
10 min	change-propagation check
5 min	title–abstract–conclusion final pass

357. Seven-day coherence cycle | 七天训练

1. Day 1: study identity cards from five papers.
2. Day 2: question–Methods handshake audits.
3. Day 3: Methods–Results orphan analysis.
4. Day 4: Results–Discussion claim-lineage audits.
5. Day 5: number/figure/table reconciliation.
6. Day 6: bilingual terminology and certainty drift.
7. Day 7: full unseen manuscript audit.

358. Twelve-week C1–C2 coherence progression

Weeks	Focus	Output
1–2	identity cards + invariant terms	study maps
3–4	section handoffs	question–method–result chains
5–6	number/outcome/analysis ledgers	audit sheets
7–8	claim lineage + causal/scope drift	repaired manuscripts
9–10	genre + bilingual transfer	multi-genre portfolio
11–12	peer-review + final submission audit	submission-ready full paper

359. Drill 1 — Question–outcome mismatch

Question:

Does intervention X produce durable learning?

Only outcome:

immediate supported task.

Repair options?

360. Model answer 1

Either add a delayed independent outcome or narrow the research question to immediate supported performance. Do not preserve the durable-learning question with inadequate measurement.

361. Drill 2 — Population drift

Methods sample:

adult volunteers from one advanced programme.

Conclusion:

This method benefits bilingual students.

362. Model answer 2

The method improved the measured outcome among the advanced adult bilingual volunteers studied; effects in other bilingual learner populations remain uncertain.

363. Drill 3 — Outcome switching

Primary performance outcome null; secondary satisfaction outcome positive. Abstract says “programme effective”. Repair the abstract conclusion.

364. Model answer 3

The programme did not produce a clear improvement in the primary performance outcome, although satisfaction was higher as a secondary outcome.

365. Drill 4 — Mechanism orphan

Discussion:

The intervention worked by reducing cognitive load.

No cognitive-load measure exists.

366. Model answer 4

Reduced cognitive load is one plausible explanation, but it was not measured directly and cannot be identified as the mechanism from the present design.

367. Drill 5 — Denominator drift

Abstract n = 120; primary analysis n = 109. What should the paper do?

368. Model answer 5

State 120 enrolled/randomised as appropriate, state 109 analysed for the primary outcome, and explain the 11-person difference through participant flow/missingness.

369. Drill 6 — Time drift

Results shows effect at four weeks. Title says “long-term”. Repair.

370. Model answer 6

Use “four-week” or another accurate horizon; reserve “long-term” for a field-appropriate extended interval.

371. Drill 7 — Adjusted/unadjusted mismatch

Methods identifies adjusted regression as primary; Abstract reports only unadjusted difference.

372. Model answer 7

Abstract should report or summarise the primary adjusted result. Unadjusted group means can remain descriptive.

373. Drill 8 — Figure title overclaim

Figure displays one-week outcome; title says “Durable Learning”.

374. Model answer 8

Rename figure to the exact outcome/time pattern, such as One-Week Independent Revision Scores by Feedback Condition.

375. Drill 9 — Terminology drift

Methods: self-efficacy scale.

Results: confidence.

Discussion: motivation.

376. Model answer 9

Use the validated construct name consistently unless confidence/motivation are separately defined and measured. Do not let stylistic variation create new constructs.

377. Drill 10 — Exploratory subgroup HARKing

A subgroup effect discovered after analysis is inserted into Introduction as a predicted mechanism.

378. Model answer 10

Restore the original Introduction. Report subgroup analysis as exploratory and present it as a hypothesis-generating finding for future confirmation.

379. Drill 11 — Chinese “效果明显” drift

One writer translates as highly effective, another as statistically significant. Repair team workflow.

380. Model answer 11

Replace the shorthand with the exact evidence state in the bilingual lexicon: for example, adjusted mean difference 5.7 points, 95% CI 2.3–9.1. Let English wording derive from the evidence, not the shorthand.

381. Drill 12 — Mixed-methods false convergence

Quant performance improves; qualitative participants report satisfaction. Discussion says both strands confirm learning.

382. Model answer 12

Performance and satisfaction are different constructs. Report the strands separately and integrate only at a higher question they genuinely share, such as intervention acceptability plus performance.

383. Drill 13 — Review scope drift

Eligibility includes only adult studies. Conclusion says intervention effective across ages.

384. Model answer 13

Restrict conclusion to adults or explicitly state that child/adolescent effectiveness remains unaddressed.

385. Drill 14 — AI benchmark drift

Model beats baseline on one benchmark. Abstract says “more capable AI”.

386. Model answer 14

State the benchmark-specific performance advantage and preserve uncertainty about broader capability.

387. Drill 15 — Feasibility/effectiveness drift

Pilot meets recruitment targets and shows a promising outcome trend. Conclusion says intervention effective.

388. Model answer 15

Conclude feasibility. Label outcome trends exploratory and reserve effectiveness for an appropriately powered comparative study.

389. Drill 16 — Professional-evaluation drift

Parent satisfaction high; learner outcomes unchanged; report calls programme successful.

390. Model answer 16

Define success by dimension: high acceptability, no clear learner-outcome improvement. Do not merge dimensions into one global success score unless criteria were prespecified.

391. Drill 17 — Protocol deviation

Primary outcome changed before analysis because instrument failure was discovered after collection. What should manuscript do?

392. Model answer 17

Report the original outcome, the reason for change, when the decision occurred relative to outcome analysis, the replacement outcome and any updated protocol/analysis plan.

393. Drill 18 — Supplement contradiction

Main text says result robust; supplement shows direction reverses under a reasonable sensitivity analysis.

394. Model answer 18

Revise the main text. Robustness claim is false. Discuss sensitivity to analytic assumptions explicitly.

395. Drill 19 — Title/Conclusion mismatch

Title says “causes”; Conclusion says association only.

396. Model answer 19

Downgrade title to association language unless design supports causality.

397. Drill 20 — Conclusion reopens rejected mechanism

Discussion concludes mechanism unresolved. Conclusion says mechanism explains benefit.

398. Model answer 20

Restore unresolved status in Conclusion. Conclusion inherits the Discussion’s surviving claim, not the writer’s preferred story.

399. The 100-point manuscript-coherence rubric

Dimension	Points	Mastery evidence
Research-question identity	10	same question survives across aim, design, result and conclusion
Construct identity	10	constructs/operationalisations remain explicit and distinct
Population/sample identity	10	target, enrolled, analysed and generalised populations reconcile
Outcome/time hierarchy	15	primary/secondary/exploratory and time points remain stable
Methods–Results lineage	15	no orphan primary methods or results; analyses match reporting
Numbers/displays consistency	10	n, estimates, units, tables and figures reconcile
Discussion claim control	10	interpretation anchored in results; mechanism/scope calibrated
Conclusion/Abstract alignment	10	front door and exit tell same evidence story
Temporal transparency	5	planned vs exploratory/deviation status visible
Bilingual terminology precision	5	Chinese/English terms do not create construct or certainty drift

400. What 90–100 looks like | C2-ready manuscript

A reader can move between sections without relearning what the study is. Every major claim has a traceable parent chain. Primary outcomes keep their priority. Numbers reconcile. Terminology is stable. Exploratory discoveries are welcomed without being rewritten as predictions. Abstract, Discussion and Conclusion differ in function but agree in evidence state.

401. What 75–89 looks like | Strong C1 manuscript

The study identity is stable, but one local drift remains: perhaps a broad title, inconsistent terminology, an unexplained denominator or a Discussion sentence that moves too far from the measured construct.

402. What 60–74 looks like | Sections are individually competent, manuscript is unstable

This is the classic advanced-writer trap. Each section sounds professional, but outcome hierarchy, construct names or claim strength drift across the paper. The repair requires horizontal auditing rather than sentence polishing.

403. Below 60 | The manuscript contains multiple research identities

The question, measurement, result and final claim do not describe one coherent study. Return to the identity card and reconstruct the paper before style revision.

404. Team self-marking protocol | Evidence for every rubric score

To award points, quote or point to:

• research question statement;
• primary outcome definition;
• primary analysis;
• primary result;
• Discussion primary interpretation;
• Conclusion primary claim;
• sample-flow explanation;
• terminology map.

405. Final unseen-paper benchmark | Hide the paper’s Abstract and Conclusion

Select an unfamiliar article. Read Introduction, Methods, Results and Discussion. Then reconstruct:

1. study identity card;
2. one-sentence question;
3. population and analysed sample;
4. primary outcome/time;
5. primary analysis;
6. primary result;
7. strongest defensible Discussion claim;
8. most consequential boundary;
9. your own Abstract conclusion;
10. your own final Conclusion.

Then reveal the paper’s actual title, Abstract and Conclusion. Audit whether the published surfaces are more or less coherent than your reconstruction.

406. Final full-paper assignment | Build a coherence package

For one paper you are writing or a fictional study, produce:

1. study identity card;
2. bilingual manuscript lexicon;
3. outcome/time ledger;
4. analysis ledger;
5. sample-flow ledger;
6. figure/table map;
7. six-surface coherence sheet;
8. claim-lineage map for five major claims;
9. protocol/registry deviation note if applicable;
10. Title;
11. Abstract;
12. final Introduction paragraph;
13. primary Methods paragraph;
14. primary Results paragraph;
15. Discussion opening;
16. Conclusion;
17. 100-point self-score with evidence.

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

• Nature Cell Biology | Methodical about Methods — methods, design, analysis, reproducibility and conclusion support as one manuscript system.
• Nature Medicine | CONSORT 2025 Statement — clear transparent reporting, alignment with protocol/analysis plans and explicit reporting of changes.
• EQUATOR Network | CONSORT 2025 — updated reporting framework for randomised trials.
• EQUATOR Network | Reporting Guidelines Library — design-specific reporting requirements across research genres.
• Nature Medicine | TRIPOD-LLM — section-spanning reporting guidance for studies using large language models.
• Nature Methods | Content Types — examples of manuscript architecture across Introduction, Results, Discussion and Methods.
• Nature Computational Science | On Writing Accessible Computational Science Papers — accessible expert writing across core manuscript sections.
• University of Manchester Academic Phrasebank — functional language across introductions, methods, results, discussion, cautious claims and conclusions.

408. Canonical eduKate research-writing route | Internal learning route

• Lesson 016 | Abstract Compression
• Lesson 017 | Research Introduction
• Lesson 018 | Methods for Reproducibility
• Lesson 019 | Results Reporting
• Lesson 020 | Discussion Without Overclaiming
• Lesson 021 | Conclusion Closure

409. SEO language map | 本课自然覆盖的搜索意图

This lesson naturally serves readers searching for research paper coherence, IMRaD structure, how to write a full research paper, align introduction methods results discussion, research question methods alignment, outcome switching, manuscript consistency, claim lineage, cross-section consistency, title abstract conclusion consistency, methods results mismatch, research paper audit, manuscript checklist, protocol manuscript consistency, research paper terminology, academic writing coherence, C1 academic writing, C2 academic English, academic English for Chinese speakers, English for Mandarin speakers, 研究论文结构, 论文各部分一致性, Introduction Methods Results Discussion 一致, 研究问题与方法一致, 研究结果与结论一致, 学术英语论文 and 中文母语学术英语.

410. Final quality gate | 20 questions before submission

1. Can the study be described in one stable sentence?
2. Does the title describe the same study?
3. Does the abstract preserve primary outcome hierarchy?
4. Does the Introduction ask a question the Methods can answer?
5. Does every promised construct have an operationalisation?
6. Does the population label remain stable?
7. Do sample counts reconcile?
8. Are primary/secondary/exploratory outcomes explicit?
9. Are time points consistent?
10. Does every major Result have a Method parent?
11. Does every primary Method produce a reported Result?
12. Do figures/tables agree with prose?
13. Do protocol/registry and manuscript differences have explanations?
14. Are exploratory findings labelled?
15. Does Discussion interpret only reported evidence?
16. Do limitations change scope/claim strength?
17. Does Conclusion inherit the surviving Discussion claim?
18. Do title, abstract and conclusion agree on causality and scope?
19. Do Chinese/English technical terms preserve the same constructs?
20. Can every major claim be traced back through result → analysis → measure → question?

411. The final principle | Full-paper coherence is research integrity made visible

A coherent manuscript is not one where every section sounds similar. It is one where every section tells the truth about the same research object from a different functional angle.

Introduction tells why the object matters.

Methods tells how it was made observable.

Results tells what was observed.

Discussion tells what the observations can mean.

Conclusion tells what survives.

整篇论文的一致性，不是每一部分“写得像”，而是每一部分从不同 job 出发，对同一个 research object 保持诚实。

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

You can take an unfamiliar manuscript and identify:

• its stable research identity;
• where that identity first drifts;
• whether numbers and outcomes reconcile;
• whether a claim has a complete lineage;
• whether exploration has been rewritten as prediction;
• whether bilingual terminology changes constructs;
• which upstream repair will fix the downstream manuscript.

At that point, you are no longer editing sections. You are controlling the integrity of a whole research argument.

---

Next lesson reserved | 下一课

EDKS-ADV-ZH-0023 · Lesson No.023 · 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

The next lesson will focus on revision after external feedback: classify reviewer requests, distinguish scientific corrections from stylistic preferences, add analyses without rewriting history, propagate changes across dependent sections, write response letters and preserve the paper’s research identity through multiple revision rounds.

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