Series ID: EDKS-ADV-ZH-0021 · How to Learn Advanced English (Chinese Edition) · Lesson No.021 · C1 → C2

Write a Conclusion That Closes the Argument Without Making It Bigger | 把研究结论写到真正收束，而不是最后再扩大主张

A research conclusion is not the place where the writer suddenly becomes more certain, more ambitious or more dramatic. It is the place where the paper’s earned answer is compressed into its cleanest final form: what the study now supports, why that matters, where the claim stops, and what genuinely remains open.

Research Conclusion 不是“最后再总结一遍”，也不是“最后再把意义拔高一点”。真正成熟的 Conclusion 必须把 Introduction 提出的问题、Results 提供的 evidence、Discussion 完成的 interpretation、limitations 与 scope 一起压缩成一个最终能够站住的 answer。它必须关闭 argument，而不是在最后一段重新打开一个更大的 argument。

For Mandarin-speaking C1–C2 writers, the final paragraph is dangerous because Chinese academic habits such as “综上所述”, “由此可见”, “研究证明”, “具有重要意义”, “值得广泛推广” and “为未来研究提供了重要参考” can be transferred into English as formulaic certainty or vague significance. The language may sound formal, but the inference can silently become stronger than the evidence. This lesson therefore treats conclusion writing as final claim control, not as decorative closure.

The operating sequence is:

question → earned answer → contribution → boundary → implication → next open question → stop.

问题 → 已经赚到的答案 → contribution → 边界 → implication → 下一步真正未解决的问题 → 停止。

---

Part I — What a research conclusion is actually for | 第一部分：Conclusion 到底在做什么

1. Why this lesson comes after Discussion | 为什么 Conclusion 必须在 Discussion 之后学

Lesson 020 taught the hard interpretive work: compare explanations, weigh literature, expose limitations, protect null results, bound generalisation and decide what the evidence can support. Conclusion does not redo that work. It inherits the final state produced by that work.

If Discussion ends with five unresolved uncertainties, Conclusion cannot erase them. If Discussion retreats from a causal claim to an association claim, Conclusion cannot restore causality. If Discussion limits the result to one population and time horizon, Conclusion cannot suddenly write about “all learners”, “education systems” or “society”.

2. Conclusion is the final compression gate | 最后的 compression gate

The full paper contains more information than the reader can carry away. The Conclusion decides what must survive compression.

A strong Conclusion preserves:

• the primary research question;
• the strongest defensible answer;
• the contribution relative to prior knowledge;
• the most important boundary or uncertainty;
• the implication that follows from the evidence;
• the next question that remains genuinely open.

3. Manchester’s two core functions | Looking back + final comment

The University of Manchester Academic Phrasebank describes conclusions as generally doing two jobs: bringing together what the text has established and providing a final comment or judgement, sometimes including future directions.

University of Manchester Academic Phrasebank | Writing Conclusions

This gives us a useful balance. Conclusion must look backward enough to consolidate the evidence, but forward enough to explain what the completed argument now means.

4. Purdue’s big-takeaway test | Reader already knows the paper

Purdue Writing Lab treats the conclusion as the place where the author summarises the paper knowing that the reader has already read it. That changes the writing task. You do not need to rebuild background, retell methods or reproduce every result.

Purdue Writing Lab | Active Reading as a Graduate Student

The reader wants the big takeaways and the final state of the argument.

5. Nature’s higher-level abstraction | Conclusion should not be a body summary

Nature’s Scitable guidance distinguishes Conclusion from mere summary: the conclusion should state the most important outcome at a higher level of abstraction and reconnect the finding to the need or motivation introduced at the beginning.

Nature Scitable | Scientific Papers

This is one of the most important principles in this lesson:

Conclusion compresses upward, not backward.

Conclusion 不是向后抄，而是向上抽象。

6. “Higher level” does not mean “bigger claim” | Abstraction ≠ inflation

Suppose Results show that one structured-feedback condition improved one-week independent revision in a specific sample.

Higher-level abstraction:

The study provides evidence that feedback design can influence near-term transfer beyond the supported task.

Inflated claim:

Structured feedback transforms how learners become independent thinkers.

The first moves from a concrete estimate to its conceptual contribution. The second moves beyond the construct, population, duration and mechanism.

7. Conclusion is not the Abstract again | Conclusion ≠ Abstract

Lesson 016 | Compress a Full Study Into an Abstract Without Distorting It

Abstract is a self-contained front-door representation. It helps readers decide what the paper is and what happened. Conclusion is an exit state for a reader who already understands the evidence and reasoning.

Abstract often includes:

• background/problem;
• method/design;
• key result;
• main interpretation.

Conclusion usually does not need to repeat all four.

8. Conclusion is not Discussion again | Conclusion ≠ Discussion

Discussion expands interpretation. Conclusion compresses the interpretation that survived Discussion.

Discussion may spend several paragraphs on:

• three competing mechanisms;
• four literature relationships;
• limitations;
• unexpected findings;
• future tests.

Conclusion may reduce all of this to:

The study supports a near-term transfer benefit but does not establish the mechanism or durability; component-controlled delayed replication is therefore the next critical test.

9. Conclusion is not a Results list | Do not replay every number

Weak:

Group A scored 74.3, Group B scored 68.1, p = .003, and the confidence interval was 2.4–9.6.

Conclusion-level:

The structured condition produced a clear near-term revision advantage.

Keep numbers only when they are necessary to define the final claim, decision threshold or practical magnitude.

10. Conclusion is not a second Introduction | Do not rebuild background

If the Introduction spent five paragraphs establishing the gap, the Conclusion does not need to retell the field history.

Instead:

The study addresses the gap by showing…

or:

These findings narrow the unresolved question to…

11. Conclusion is not a new literature review | No late citation parade

A major new theory or literature body introduced only in the Conclusion has arrived too late. The reader has had no opportunity to see it tested against the paper’s evidence.

Minor contextual citations can be appropriate in some disciplines, but the main architecture should not depend on new sources.

12. Conclusion is not a marketing paragraph | Avoid hype at the exit

Words such as:

• groundbreaking;
• revolutionary;
• transformative;
• unprecedented;
• game-changing;
• profound;

should not substitute for a specific contribution.

Instead of:

This groundbreaking study has profound implications.

write:

The study provides the first delayed independent-transfer test in this programme, separating supported performance from later unaided revision.

13. Conclusion is not a victory speech | Hypothesis can fail and paper can still conclude well

A null or mixed result can produce a strong conclusion.

The intervention did not improve the primary outcome, narrowing the plausible benefit to self-reported confidence rather than demonstrated performance.

This is knowledge.

14. Conclusion is not where uncertainty disappears | Uncertainty survives compression

Compression removes detail, not epistemic boundaries.

If Discussion says:

mechanism uncertain; long-term effect uncertain; single-site transportability uncertain

Conclusion cannot write:

The intervention is an effective solution.

15. Conclusion is not required in every article | Genre matters

Manchester notes that separate conclusions may be optional in research articles where Discussion already consolidates findings and implications. Some journals end with a Discussion; others require a distinct Conclusion; dissertations and essays often expect one.

Always inspect target-venue conventions.

16. Even without a heading, conclusion work still exists | Final closure function

A paper may have no heading labelled “Conclusion” but still needs a concluding move. The final Discussion paragraph may perform it.

17. The minimum viable conclusion | What must survive if space is severe?

At minimum:

1. answer the main question;
2. state the contribution;
3. state the most important boundary;
4. state the next implication or unresolved question.

18. The closing-loop principle | Introduction opens; Conclusion closes

Lesson 017 | Write an Introduction Until the Research Question Becomes Necessary

Introduction:

We do not know whether supported improvement survives when assistance is removed.

Conclusion:

The present trial shows that part of the improvement survives one week without direct assistance, while longer-term persistence remains unresolved.

This is a closed loop.

19. A conclusion should answer the question the paper actually asked | No last-minute topic change

If the paper asks about short-term independent revision, the Conclusion should not suddenly become a manifesto about educational inequality, lifelong learning or national curriculum reform unless those issues were genuinely part of the study and Discussion.

20. A conclusion can narrow the original question | Evidence is allowed to discipline ambition

Introduction may ask:

Does X create durable independent capability?

Data may support only:

X improves near-term independent performance.

Conclusion should state the narrower answer.

21. The final thesis must be evidence-responsive | Final thesis may differ from initial hypothesis

The hypothesis is what you thought might be true before the evidence. The final thesis is what the completed study supports after analysis.

22. “Closing” means no major unresolved logical debt | Argument debt

A paper is not closed if the final paragraph leaves the reader asking:

• But what was the primary outcome?
• Was the result causal or associative?
• Who does this apply to?
• Did the null result matter?
• Why does this contribution matter?

23. But closure does not mean pretending all questions are solved | Open questions can be part of closure

A strong Conclusion can close the current paper precisely by naming what remains open.

The study resolves the near-term transfer question but leaves mechanism and six-month durability for future tests.

24. The earned-answer principle | “Strongest defensible” not “strongest imaginable”

The best final claim is the strongest statement that survives:

• design constraints;
• measurement constraints;
• uncertainty;
• counterevidence;
• alternative explanations;
• scope limits.

25. Do not weaken a strong result out of habit | Caution can also distort

If a well-designed randomised study with a precise estimate supports a clear short-term effect, writing:

There might possibly perhaps be some indication of a potential effect.

is not sophisticated. It understates the evidence.

Write:

The intervention improved the measured one-week outcome in this sample.

26. Final claims should be dimension-specific | Strength can differ across dimensions

You may conclude:

• high confidence in direction;
• moderate confidence in magnitude;
• low confidence in mechanism;
• low confidence in long-term durability.

Conclusion can preserve this multidimensional state in a compact form.

27. The four-sentence conclusion architecture | A robust default

Sentence 1 — Answer: What did the study establish?

Sentence 2 — Contribution: What does this add to the field/problem?

Sentence 3 — Boundary: What is the most consequential uncertainty or limit?

Sentence 4 — Next implication: What action, test or question now follows?

28. Example four-sentence conclusion | 示例

Structured action-oriented feedback improved near-term independent revision, with the clearest advantage in evidence integration. This extends earlier supported-task work by showing that part of the benefit survives when direct feedback is removed. The study does not isolate the mechanism or establish long-term durability beyond the measured follow-up. Component-controlled delayed replication is therefore the next critical test before routine implementation claims are made.

29. The five-layer architecture | When the paper needs more nuance

1. research question;
2. earned answer;
3. contribution;
4. boundary;
5. implication/future direction.

30. Restating the research aim is optional, not automatic | Avoid ritual openings

Weak ritual:

The aim of this study was to investigate…

If the paper is long and the research question is complex, a short restatement may orient the reader. If the question is obvious, begin with the answer instead.

31. Answer-first conclusions are often stronger | Message before process

Instead of:

This study investigated the effects of X on Y.

write:

X improved Y under the tested conditions but did not produce a clear delayed effect.

32. Contribution should be relational | Contribution relative to what was missing?

A contribution sentence should answer:

What can the field now say that it could not say before this study?

33. Contribution type 1 — New evidence | 新 evidence

The study provides the first delayed unaided test in this sample.

34. Contribution type 2 — Boundary condition | 找到边界

The benefit was concentrated among novice learners, identifying baseline proficiency as a candidate boundary condition.

35. Contribution type 3 — Mechanism discrimination | 区分 explanations

Holding contact time constant reduced the difference, weakening the explanation that feedback structure alone drove the effect.

36. Contribution type 4 — Methodological advance | 方法贡献

The study introduces an external validation protocol that separates benchmark fit from cross-site transportability.

37. Contribution type 5 — Negative evidence | Null results can contribute

The absence of delayed benefit narrows claims about durable learning and redirects attention toward transient performance support.

38. Contribution type 6 — Synthesis | 整合贡献

The review shows that apparently conflicting studies separate when follow-up duration and baseline proficiency are considered together.

39. Contribution type 7 — Measurement clarification | 测量贡献

The study shows that confidence and objective performance can move independently, limiting use of self-confidence as a proxy for competence.

40. Contribution type 8 — Practical feasibility | 实施贡献

The pilot establishes that the protocol can be delivered within routine appointment time, although effectiveness remains uncertain.

41. Avoid generic “fills a gap” language | Name the gap that changed

Weak:

This study fills an important gap in the literature.

Stronger:

The study supplies the delayed independent task missing from earlier evaluations that measured only supported performance.

42. Avoid generic “adds to knowledge” language | What knowledge?

State the changed proposition.

43. Avoid generic “important implications” | What implication?

Weak:

The findings have important implications for teaching.

Stronger:

The pattern suggests that feedback studies should measure transfer after support removal rather than infer learning from assisted-task improvement alone.

44. The boundary sentence is not optional when scope matters | Claim stops somewhere

Possible boundary dimensions:

• population;
• setting;
• time;
• outcome;
• mechanism;
• dose;
• implementation conditions;
• measurement validity.

45. One limitation can dominate the conclusion | Prioritise

If the study’s biggest problem is no delayed follow-up, the final boundary should probably mention durability—not a minor procedural limitation.

46. Boundary does not mean weakness | It makes the claim useful

The intervention improved one-week performance but durability beyond four weeks remains untested.

This tells the reader exactly what they can safely carry forward.

47. Final implication must follow from the final claim | No leap after the boundary

If the final claim is narrow, the implication must respect it.

Claim:

Short-term supported performance improved.

Valid implication:

Future evaluations should include delayed independent tasks.

Invalid leap:

Schools should adopt the intervention nationally.

48. Future work should close the biggest uncertainty | Not a generic wish list

If mechanism is the key uncertainty:

Future work should isolate action structure from planning time.

If durability is the key uncertainty:

A six-month unsupported follow-up is the next test.

49. Do not end with “more research is needed” | Specify the missing evidence

Weak:

More research is needed.

Strong:

A multi-site delayed replication with matched contact time is needed to determine whether the observed benefit is durable and attributable to feedback structure rather than additional planning.

50. End with the next state of knowledge | Knowledge frontier

A good final sentence often tells the reader what the problem has become after this paper.

Before study:

Does structured feedback transfer?

After study:

Near-term transfer appears plausible; durability and mechanism are now the decisive questions.

51. Conclusion as versioning | The question changes version

A research paper begins with Question v1.0. Evidence and Discussion produce Answer v1.0 plus Uncertainty v2.0. Conclusion should state that new state cleanly.

52. The stop condition | A conclusion needs to know when to stop

When the paper has stated:

• the answer;
• the contribution;
• the boundary;
• the next implication;

stop. Another paragraph often weakens closure by reintroducing background, citations or ambition.

53. Nature Physics: conclusions should give perspective, not lazy repetition | Perspective after evidence

Nature Physics has argued against conclusions that simply summarise the paper, emphasising instead that a useful conclusion gives perspective on what the reader has learned.

Nature Physics | Elements of Style

54. “Perspective” still needs evidential control | Perspective ≠ speculation without limit

A perspective sentence should grow from the result:

The finding shifts the design question from whether prompts help to which prompt component produces transferable benefit.

It should not become:

This research will reshape the future of education.

55. The reader-memory test | What should remain tomorrow?

After reading the whole paper, what one or two propositions should a careful reader remember? Those propositions belong in the Conclusion.

56. The decision test | What should a practitioner or researcher do differently?

If the paper has practical relevance, specify the decision implication at the level supported by evidence.

57. The theory test | What changed in the conceptual model?

Did the study:

• support a mechanism?
• weaken one?
• identify a boundary?
• show two constructs diverge?
• leave competing models unresolved?

58. The method test | What should future studies measure differently?

A paper can conclude with methodological contribution even when substantive effect is uncertain.

59. The honesty test | Could a sceptical reader accept the final claim?

The Conclusion should be written so that someone who rejects your preferred theory can still accept that you represented the evidence fairly.

60. Part I operating rule | 最终规则

Do not ask: “How can I make the ending sound stronger?” Ask: “What is the strongest answer that survives everything the paper has already admitted?”

不要问：“最后怎样写得更有力？”要问：“在整篇文章已经承认的 evidence、counterevidence、limitations 与 uncertainty 之后，什么是还能站住的最强 answer？”

---

Part I establishes the conclusion as a closure-and-compression system. The next section will control claim size, causality, generalisation, significance, implications and future directions so that the final paragraph cannot become stronger than the completed paper.

Part II — Final claim control: causality, scope, significance and future direction | 第二部分：控制最终 claim 的大小

61. Conclusion inherits the study design | 设计决定最终动词

The final claim cannot use a causal verb stronger than the design can support. This sounds obvious, but the Conclusion is where writers often upgrade the verb because the paper now feels complete.

Observational:

X was associated with Y.

Randomised intervention:

Assignment to X improved Y under the tested conditions.

Do not let the emotional sense of “we have finished the paper” become epistemic permission.

62. A causal verb carries counterfactual meaning | Cause = what would happen otherwise

Words such as causes, leads to, produces, prevents and often improves imply that changing X would change Y, other things equal. Conclusion should use these only when the design and analysis support that counterfactual interpretation.

63. “Predicts” is not automatically causal | Statistical prediction vs causal effect

A model may predict an outcome without identifying the cause. If your paper is predictive, conclude in predictive language.

The model predicted 30-day readmission with improved external-site discrimination.

not:

The identified variables cause readmission.

64. “Explains” needs special care | Variance explained ≠ causal explanation

Regression may explain variance statistically. A theoretical mechanism explains causally. Conclusion should not blur the two.

65. “Mediates” and “moderates” are technical claims | Do not use casually

If the paper tests mediation or moderation under explicit assumptions, the final claim can reflect that. If the paper only observes subgroup differences or correlations, do not conclude that a variable “mediates” or “moderates” the effect.

66. Strong causal design still needs scope control | Causality and generalisation are different

A randomised trial may strongly support a causal effect in the studied sample while saying little about another population, setting or implementation condition.

Internal validity does not automatically buy external validity.

67. Population scope | Who was actually studied?

Conclusion nouns should preserve the population:

Weak:

Students benefit from the intervention.

Stronger:

Advanced adult bilingual learners in the studied programme showed a near-term benefit.

68. Setting scope | Where did the effect occur?

If the intervention depended on experienced instructors, small groups or high resource levels, the Conclusion should not treat the effect as setting-free.

69. Time scope | When is the claim true?

Immediate, one week, twelve weeks and one year are different claims.

The effect persisted to four weeks.

is not:

The effect is durable.

70. Outcome scope | Which outcome changed?

If evidence integration improved but grammar accuracy did not, the conclusion should not compress both into “writing improved” without qualification.

71. Construct scope | Measured proxy vs broad concept

Attendance can measure behavioural participation. It does not fully measure engagement. A confidence scale does not equal competence. A benchmark score does not equal general intelligence. Conclusion must keep the measured construct visible.

72. Dose scope | Which intervention version?

One session, six weeks and one semester are not interchangeable. If effect depends on dose, preserve it.

73. Implementation scope | Who delivered it?

Designer-led implementation may not travel directly to routine use. A conclusion about efficacy should not quietly become a conclusion about implementation effectiveness.

74. Comparator scope | Better than what?

“Effective” without a comparator can be misleading. Was the intervention better than no treatment, business as usual, an active comparison or baseline?

75. The claim-scope vector | A compact control tool

Before writing the final sentence, fill in:

Population × Setting × Time × Outcome × Comparator × Dose × Mechanism certainty.

Your final claim should not exceed this vector.

76. Example claim-scope vector | 示例

Population: advanced bilingual adults.

Setting: one high-support programme.

Time: one week.

Outcome: independent revision score.

Comparator: standard evaluative feedback.

Dose: four structured feedback sessions.

Mechanism: uncertain.

Valid conclusion:

Four sessions of structured feedback improved one-week independent revision relative to standard evaluative feedback in this advanced bilingual sample; the mechanism and longer-term durability remain unresolved.

77. Statistical significance does not deserve the final sentence by itself | p-value is not the conclusion

Conclusion should focus on effect, uncertainty and meaning. “Statistically significant” is rarely the most useful takeaway unless significance testing is itself central to the research question.

78. Practical significance needs a criterion | Meaningful compared with what?

A statistically clear effect can be practically trivial. A small effect can be practically valuable if cost is low and scale is large. Conclusion should connect magnitude to a relevant criterion.

79. Absolute and relative effects can tell different stories | Preserve both when decisions depend on them

A risk reduction from 2% to 1% is a 50% relative reduction and a one-percentage-point absolute reduction. Conclusion should use the representation that supports informed interpretation, not the most dramatic framing.

80. Precision matters | Wide intervals should sound wide

If an estimate ranges from trivial to substantial, the final claim should acknowledge magnitude uncertainty.

The direction favoured the intervention, but the effect size remains imprecise.

81. Null result ≠ no effect | Do not compress uncertainty into zero

Weak:

The intervention had no effect.

Stronger:

The study did not detect a clear effect, and the interval remains compatible with effects in either direction.

82. Equivalence requires equivalence evidence | Same is a positive claim

To conclude that two approaches are equivalent, the design must support equivalence or non-inferiority logic. Ordinary non-significance is insufficient.

83. Negative result can narrow theory | Null findings are not empty

The absence of delayed transfer weakens the claim that the intervention produced durable learning under the tested dose.

This is a substantive contribution.

84. Mixed findings require mixed conclusions | Do not force one-direction closure

If speed improves and accuracy declines, the final claim should represent the trade-off:

The system increased throughput at the cost of lower accuracy.

85. Harms belong in final compression when they change value | Benefit without harm is incomplete

If an intervention improves the target outcome but increases adverse events, the Conclusion must not present only the benefit.

86. Benefit–burden conclusions | Multi-objective decision

The programme improved completion rates but required substantially more staff time, so its practical value depends on whether the additional gain justifies the implementation burden.

87. Conclusion can state a trade-off without deciding it | Values may belong to decision-maker

The evidence can identify the trade-off. Whether the trade-off is acceptable may depend on goals and values outside the study.

88. Implication ≠ recommendation | Final paragraph must keep them separate

Implication: what the evidence changes in understanding.

Recommendation: what someone should do.

Recommendation requires additional decision criteria.

89. Recommendation strength should scale with evidence and consequence | Error cost matters

A low-cost reversible pilot can be reasonable under moderate uncertainty. A high-cost irreversible policy usually requires stronger evidence.

90. “Should” is a high-burden word | Ask what decision logic supports it

Before writing should, identify:

• goal;
• expected benefit;
• risk;
• cost;
• feasibility;
• alternatives;
• reversibility.

91. Low-risk conclusion example | Pilot recommendation

Given the modest benefit, low implementation cost and reversibility, a larger pilot is justified while durability is tested.

92. High-risk conclusion example | No premature adoption

The current evidence is insufficient to support routine adoption because the intervention’s long-term harms and effectiveness under ordinary conditions remain uncertain.

93. Theory implication | What changed in the model?

A conclusion can state:

The result supports a boundary condition.

The result weakens one mechanism.

The study leaves two mechanisms unresolved.

The evidence separates two previously conflated constructs.

94. Avoid “confirms theory” unless rivals are excluded | Compatible is often enough

If Theory A and Theory B predict the same result, Conclusion cannot declare Theory A confirmed.

95. Theory refinement can be stronger than theory victory | Boundary knowledge accumulates

The benefit was limited to lower-baseline learners, suggesting that prior proficiency may constrain when the mechanism operates.

96. Methodological implication | What should future researchers measure?

Future trials should include unsupported delayed tasks because supported-task performance alone overstates the evidence for independent learning.

97. Measurement implication | When proxy and target diverge

Because confidence increased without objective performance change, future evaluations should not treat confidence as a stand-alone proxy for competence.

98. Replication implication | When one result needs independent confirmation

Independent multi-site replication is needed to determine whether the effect survives new instructors and implementation contexts.

99. Mechanism implication | Design the next discriminating test

Matching planning time while varying prompt structure would distinguish action clarity from additional cognitive effort.

100. Durability implication | Time is the missing evidence

A six-month unaided follow-up is the next critical test.

101. Generalisation implication | Transportability test

Replication in lower-support settings is needed before routine-setting effectiveness can be inferred.

102. Scale implication | Implementation science question

The next uncertainty is not efficacy but whether the programme can be delivered consistently at larger scale.

103. Future work must inherit the biggest uncertainty | Do not list random possibilities

If the paper’s core limitation is mechanism ambiguity, ending with “future work should include larger samples” misses the problem.

104. Future work should be discriminating | A good study can change the conclusion

Ask:

What result would make us prefer Explanation A over Explanation B?

Design the next study around that contrast.

105. Future work should be feasible enough to be meaningful | Avoid impossible wish lists

“Future research should examine all populations over many years” is not a research plan. Specify the next high-value uncertainty.

106. Future work should not be self-advertisement | Avoid “we will…” unless concrete

Nature Scitable distinguishes perspectives from firm future plans. If you state a future plan, make clear whether it is an actual planned study or a broader invitation to the field.

107. The final sentence can point forward | But it must remain connected

Strong:

The next question is therefore whether the near-term gain survives prolonged unsupported use.

Weak:

Future research should explore many other interesting possibilities.

108. Ending with a question | Sometimes useful, not always

A final question can create intellectual continuity, but it should be the question produced by the paper—not a rhetorical flourish.

109. Ending with a recommendation | Use only when decision logic is present

If the paper is evaluative or policy-oriented, a recommendation may be appropriate. It should not appear as a genre habit in every conclusion.

110. Ending with a contribution | Often the cleanest close

By separating assisted performance from delayed unaided transfer, the study narrows what can legitimately be claimed as learning.

This gives perspective without opening another branch.

111. Ending with a boundary | Powerful when overclaim risk is high

The evidence supports a short-term effect under the tested conditions, not a universal or durable benefit.

112. Ending with a next test | Strong scientific closure

The remaining mechanism question can now be tested directly by matching planning effort across conditions.

113. Do not end with generic importance | “This is important” is not content

Tell the reader what changed.

114. Do not end with a quotation unless the genre supports it | Avoid borrowed drama

Research conclusions usually gain authority from evidential precision, not inspirational quotation.

115. Do not end with an unrelated global issue | No scale jump

A classroom intervention does not automatically justify a final sentence about the future of global education.

116. The final paragraph should usually shrink uncertainty, not enlarge topic scope | Narrower frontier

A paper should often end with a more precise problem than it began with.

117. The conclusion-consistency audit | Check six surfaces

Compare the final claim with:

1. research question;
2. primary outcome;
3. Results estimate;
4. Discussion interpretation;
5. limitations;
6. abstract/title.

Any mismatch is a warning.

118. Title–Conclusion consistency | High leverage

If the Conclusion says “associated with”, the title should not say “causes”. If the Conclusion says “short-term”, the title should not say “lasting”.

119. Abstract–Conclusion consistency | Front door and exit must agree

The abstract is written for a reader before the paper. The Conclusion is written after the paper. Their level of certainty and scope should still converge.

120. Part II operating rule | 最终 claim control

Every word that enlarges causality, population, time, construct, recommendation or significance adds an evidential burden. If the paper does not pay that burden, the Conclusion must not spend it.

任何把 causality、population、time、construct、recommendation 或 significance 扩大的词，都会增加 evidential burden。整篇论文如果没有支付这个 burden，Conclusion 就不能花掉这笔“证据预算”。

---

Part II keeps the final claim inside its evidence budget. Part III will address the special bilingual problem: how common Mandarin academic closing habits can become vague, inflated or unnatural when transferred directly into advanced English.

Part III — Mandarin-to-English conclusion reconstruction | 第三部分：把中文式结尾重建成 C1–C2 英语

121. The bilingual problem is inferential, not lexical | 真正的问题不是词汇

Many advanced Mandarin-speaking writers know the English words they need. The difficulty is that a familiar Chinese closing move can carry several possible levels of certainty, causality and significance, while an English equivalent may commit the writer to a narrower and stronger proposition.

Therefore the workflow is:

Chinese meaning state → evidence relation → English reconstruction.

not:

Chinese phrase → dictionary equivalent.

122. “综上所述” | In conclusion / Taken together / Overall

“综上所述” can introduce a synthesis, but In conclusion is not always necessary in a short research conclusion. Often the first sentence should simply state the earned answer.

Formulaic:

In conclusion, this study has shown that…

More direct:

Structured feedback improved near-term independent revision but did not establish long-term durability.

123. “由此可见” | Do not default to “it can therefore be seen that”

This literal pattern can sound awkward and may overstate logical necessity.

Possible reconstructions:

• Taken together, the findings suggest…
• The evidence supports…
• The study therefore narrows the conclusion to…
• These results indicate…

124. “研究证明” | Proves is usually too strong

In empirical research, proves is rarely the default.

Choose based on design:

provides evidence that

supports the conclusion that

shows under the tested conditions that

demonstrates only when the evidence relation is direct and strong.

125. “这说明” | What kind of “说明”?

It may mean:

• directly shows;
• suggests;
• is consistent with;
• supports;
• helps explain.

The Conclusion should choose the exact relation.

126. “因此” | Therefore requires real inferential continuity

Do not use therefore to jump from a small short-term result to a broad recommendation.

Weak:

The intervention improved one-week scores; therefore schools should adopt it.

Repair:

The one-week improvement justifies further implementation testing, while long-term benefit and routine feasibility remain uncertain.

127. “可见该方法有效” | Effective is not a free-standing property

Ask:

• effective for which outcome?
• against which comparator?
• over what time?
• in which population?

Repair:

The method improved the prespecified short-term outcome relative to standard practice in the studied sample.

128. “具有重要意义” | Important significance is not self-explanatory

English academic readers need to know what changed.

Weak:

The findings have important theoretical and practical significance.

Stronger:

The findings separate immediate supported performance from delayed independent transfer, narrowing what can be claimed as learning.

129. “具有理论意义” | Name the theoretical change

Did the study:

• support a boundary?
• weaken a mechanism?
• distinguish two models?
• show constructs diverge?

Say which.

130. “具有实践意义” | Name the decision consequence

The result suggests that programme evaluations should include delayed unaided tasks before assisted gains are treated as durable learning.

131. “值得推广” | Worth promoting requires decision criteria

Do not translate directly as is worth promoting unless cost, risk, feasibility and generalisation support that recommendation.

Possible repair:

The low-cost approach warrants larger routine-setting trials before broader adoption is considered.

132. “值得广泛应用” | Broad application is a very strong claim

A small single-site study usually supports further testing, not broad deployment.

133. “为实际工作提供参考” | Too vague in English

What should practitioners do differently?

Weak:

The study provides a useful reference for practice.

Stronger:

The study identifies delayed independent performance as a necessary outcome when evaluating feedback programmes.

134. “为未来研究提供参考” | Future research needs a specific design implication

Future studies should match planning time across conditions to isolate whether action structure itself produces the observed benefit.

135. “有待进一步研究” | Name the unresolved question

Do not end with a generic phrase.

Whether the one-week benefit persists for six months remains unresolved.

136. “需要进一步验证” | Validation of what, by what evidence?

Possible meanings:

• replication;
• external validation;
• mechanism test;
• longer follow-up;
• instrument validation.

Choose exactly.

137. “结果表明该模型具有较好效果” | Better by which metric?

Repair:

The model improved external-site discrimination while calibration changed little.

138. “总体来看” | Overall can erase mixed outcomes

Weak:

Overall, the intervention performed well.

Stronger:

The intervention improved speed and user satisfaction, while accuracy remained unchanged.

139. “基本达到预期” | Compare with prespecified criterion

The study met the prespecified threshold for short-term improvement but not the delayed-transfer criterion.

140. “效果较为理想” | Replace judgement with metric

The observed effect exceeded the prespecified minimum practical difference.

141. “结果较稳定” | Stable across what?

The direction was consistent across two samples and three prespecified sensitivity analyses.

142. “结论具有可靠性” | Reliability needs a dimension

Possible repair:

The primary direction was robust across alternative missing-data specifications, although external validity remains limited by the single-site sample.

143. “具有一定局限性” | Never leave “certain limitations” vague

Write:

The short follow-up limits conclusions about durability.

or:

Volunteer recruitment limits representativeness.

144. “在一定程度上支持” | State which dimension is supported

The result supports the predicted direction but leaves the magnitude uncertain.

145. “不能完全说明” | What exactly remains unestablished?

The result establishes association but not the proposed causal mechanism.

146. “不能排除” | Cannot rule out is not probability

Residual confounding cannot be excluded.

This means the possibility remains open. It does not mean residual confounding is likely.

147. “可能由于” | Possible explanation vs supported explanation

One possible explanation is…

or:

The pattern is consistent with…

Use the second when there is some positive fit, not merely absence of exclusion.

148. “很可能” | Do not translate to “very likely” without probability evidence

Often:

This explanation is plausible because…

is more accurate.

149. “应该” | Expectation, recommendation or requirement?

Chinese “应该” can cover several functions. English must separate them:

is expected to — prediction.

should be tested — recommendation.

must meet — requirement.

150. “必须” | Must needs authority or logical necessity

Do not use must just to make the ending sound decisive.

151. “显著” | Significant has two meanings

Chinese “显著” may mean noticeable or statistically significant. English Conclusion should distinguish:

• statistically significant;
• substantial;
• large;
• clear;
• meaningful.

152. “明显” | Clearly/obviously are often unnecessary boosters

Instead of:

Clearly, the programme was more effective.

write the evidence state.

153. “一定程度” | Avoid vague partiality

Say whether the limitation applies to:

• magnitude;
• duration;
• scope;
• mechanism;
• generalisation.

154. “基本可以认为” | Writer permission is not evidence

Replace with:

The evidence supports the conclusion that…

155. “充分说明” | Sufficient for which claim?

The converging results provide strong evidence for the short-term effect, but not for long-term durability.

156. “进一步说明” | What did the additional evidence change?

The second sample reproduced the direction, increasing confidence that the pattern is not unique to the first cohort.

157. “相互印证” | Converging evidence can still share bias

Behavioural and interview findings converged on a similar pattern, although both were collected in the same intervention context.

158. “从某种意义上” | Often a sign the claim is underspecified

Ask what exact sense is meant. Replace rhetorical vagueness with the relevant dimension.

159. “具有一定价值” | Value for what job?

The study contributes a delayed independent outcome that previous evaluations lacked.

160. “具有一定创新性” | Innovation should be specific

Do not praise novelty generically.

The design isolates supported and unsupported performance within the same participants.

161. “首次发现” | First claims require literature confidence

If you cannot establish priority reliably, avoid absolute “first” language.

Safer:

To our knowledge, this is among the first studies to…

or simply state the contribution without priority.

162. “填补空白” | Gap-filling must name the gap

The study addresses the missing delayed-transfer evidence in a literature dominated by immediate supported tasks.

163. “丰富了理论” | How did the theory change?

The findings add a proficiency boundary to the existing model.

164. “完善了模型” | Improvement needs a defined modification

Adding implementation intensity improved external calibration across the two sites.

165. “有助于理解” | What does the reader now understand better?

The comparison clarifies why immediate performance gains can occur without durable transfer.

166. “提供新的视角” | Perspective should be identifiable

The results shift evaluation from assisted output quality to the distinction between assisted performance and independent capability.

167. “具有指导意义” | Guidance needs an actionable implication

The findings support using delayed unaided tasks as a decision gate before programmes claim independent learning.

168. “有利于” | Beneficial for what outcome?

The intervention reduced completion time.

Do not write is beneficial when benefits and costs are mixed.

169. “不容忽视” | Avoid rhetorical urgency without evidence

If a limitation matters, state its consequence.

170. “值得注意” | Explain why it changes the conclusion

Weak:

It is worth noting that the delayed effect was smaller.

Stronger:

The smaller delayed effect weakens the evidence for persistence.

171. “未来具有广阔前景” | Avoid prediction-as-praise

Research Conclusions should not forecast success without evidence.

Possible repair:

The approach is sufficiently promising to justify larger routine-setting evaluation.

172. “对未来研究具有重要启示” | Name the research design change

Future studies should separate prompt structure from additional planning time to identify the active component.

173. “可以推广到其他领域” | Cross-domain transfer needs evidence

Do not infer transportability from conceptual similarity alone.

Whether the effect extends to other domains remains an empirical question.

174. “具有普遍意义” | Universal significance is almost never free

State the broader principle only if evidence and reasoning support it, then preserve the empirical scope of the study.

175. “研究目的基本实现” | Process-centred conclusion is weak

The reader cares more about what was learned than whether the researchers completed their plan.

Replace:

The study achieved its objectives.

with the actual answer.

176. “证明了研究假设” | Hypothesis support is not absolute proof

The primary result supported the preregistered hypothesis.

Then state boundaries.

177. “未能验证假设” | Null result should not be framed only as failure

The primary outcome did not support the hypothesised effect, narrowing the plausible benefit to the secondary self-report outcome.

178. “未来还需深入研究” | Deep research is not a design

Name the next evidence required.

179. The bilingual reconstruction checklist | 中文→英语 Conclusion 五步

1. Remove formulaic praise. 删除“重要、显著、广阔、创新”等没有 evidence job 的词。
2. Identify the evidence relation. 是 shows / supports / suggests / consistent with / cannot rule out 哪一个？
3. Restore scope. population / setting / time / outcome / comparator。
4. Name the contribution. 不是“有意义”，而是“改变了哪一个 proposition”。
5. Name the next unresolved test. 不写 generic “more research”。

180. Part III operating rule | 不要翻译 conclusion，要重建 conclusion

A strong bilingual writer does not ask, “How do I translate 综上所述、由此可见、具有重要意义?” The writer asks, “What evidence state is this Chinese sentence trying to express, and what English claim matches that state exactly?”

高级双语写作不是把“综上所述、由此可见、具有重要意义”翻成英文。真正的工作是：先确定中文想表达的 evidence state，再用最匹配的 English claim 重建。

---

Part III repairs conclusion language at the level of inference. Part IV will now apply the system to full worked conclusions across quantitative, qualitative, mixed-methods, computational, review and professional-report genres.

Part IV — Full worked conclusions across research genres | 第四部分：不同研究类型的完整 Conclusion

181. Why genre transfer matters | 一个模板不能解决所有 Conclusion

The core jobs of conclusion writing remain stable, but the visible form changes by research genre. A randomised trial, qualitative study, mixed-methods paper, systematic review, computational benchmark, engineering validation report and professional evaluation should not all end in identical language.

182. Worked case A — Randomised education trial | 教育随机试验

Fictional teaching case.

120 advanced bilingual learners are randomised to structured revision prompts or standard feedback. The primary one-week independent-revision outcome favours structured prompts by 5.8 points, with a reasonably precise interval. At 12 weeks, the estimated difference is 1.4 points and imprecise. Planning time is higher in the structured condition.

183. Weak conclusion for case A | 弱版本

In conclusion, structured feedback is clearly an effective method that improves students’ writing ability. This study proves that giving learners clearer guidance helps them become more independent and has important implications for teaching. Schools should use this method widely, although more research is needed.

184. Diagnose the weak conclusion | 五个问题

• “effective method” lacks time/outcome scope.
• “writing ability” inflates one revision outcome into a broad construct.
• “proves” overstates mechanism.
• “schools should use widely” outruns routine-setting evidence.
• “more research is needed” fails to identify the decisive uncertainty.

185. Strong conclusion for case A | 强版本

Structured revision prompts improved one-week independent revision relative to standard feedback in this advanced bilingual sample, with the clearest advantage in evidence integration. This extends earlier supported-task work by showing that part of the benefit survives after direct feedback is removed. The smaller and imprecise 12-week estimate leaves durability unresolved, and the additional planning time prevents the study from isolating prompt structure as the unique mechanism. A delayed component-controlled replication is therefore the next critical test before broader implementation claims are made.

186. Why this works | 四层 closure

The conclusion answers the primary question, identifies the contribution, preserves the delayed uncertainty, names the mechanism problem and converts that uncertainty into a specific next study.

187. Worked case B — Observational workplace study | 职场观察研究

Fictional teaching case.

Twelve firms choose whether to adopt a four-day workweek. Adopter firms show higher productivity and self-reported wellbeing after three months. Adopter firms also had greater baseline autonomy and lower turnover, and two introduced new project-management software.

188. Weak conclusion for case B

The four-day workweek increases productivity and wellbeing and should therefore be adopted by modern companies.

189. Strong conclusion for case B

Voluntary adoption of a four-day schedule was associated with higher short-term productivity and reported wellbeing in this group of firms. Because programme adoption was not random and adopter firms differed in baseline autonomy, turnover and concurrent process changes, the study does not establish a clean causal schedule effect. The volunteer sample also limits transportability to organisations with different staffing or service constraints. Stronger quasi-experimental or randomised evidence is needed before the observed association is treated as a general productivity effect.

190. Genre lesson from case B | Observational conclusions should not become policy slogans

The paper can conclude strongly about the observed association and clearly about its causal limits at the same time.

191. Worked case C — Qualitative interview study | 质性研究

Fictional teaching case.

Thirty multilingual postgraduate writers are interviewed about supervisor feedback. Three themes recur: action clarity, emotional safety and overload. Most participants value specific action-oriented comments, but highly detailed feedback sometimes produces paralysis among writers who cannot prioritise revisions.

192. Weak qualitative conclusion

Students prefer detailed feedback and need supportive supervisors.

193. Strong qualitative conclusion

Participants experienced feedback as most useful when it combined clear revision actions with enough prioritisation to prevent overload. This complicates a simple “more feedback is better” account: detail supported writers who could organise it, but became constraining when learners lacked a stable way to rank competing revision demands. The study therefore shifts the design question from feedback quantity to the interaction between action clarity and prioritisation. Because the sample comprised volunteer postgraduate writers in one institutional context, the transferability of these themes to earlier-stage or compulsory settings remains to be tested.

194. Qualitative conclusion contribution | Pattern, mechanism candidate, context

The conclusion does not turn themes into population frequencies or causal effects. It states the conceptual pattern and its boundary.

195. Worked case D — Mixed-methods study | 混合方法研究

Fictional teaching case.

A writing intervention improves rubric scores moderately. Interviews show most participants found the prompts clearer, but a subgroup felt the prompts restricted originality. Log data show faster revision completion.

196. Weak mixed-methods conclusion

The intervention improved performance because students found the prompts clearer.

197. Strong mixed-methods conclusion

The intervention produced a moderate performance advantage and shorter revision times, while participant accounts identified action clarity as a plausible contributor to the benefit. At the same time, reports of restriction indicate that the same structure may become constraining for writers who already possess stable revision strategies. The integrated evidence therefore supports a benefit–constraint trade-off rather than a uniformly positive mechanism. Future studies should manipulate prompt structure directly and examine whether baseline revision expertise moderates the effect.

198. Mixed-methods conclusion job | Integrate, do not stack

The final paragraph should state what becomes visible only when quantitative and qualitative strands are considered together.

199. Worked case E — Systematic review | 系统综述

Fictional teaching case.

A review includes 28 vocabulary retrieval-practice studies. Short-term pooled effects are positive. Delayed effects are smaller and heterogeneous. Only five studies follow learners beyond eight weeks. Three advanced-learner studies show little average benefit.

200. Weak review conclusion

Retrieval practice is an effective vocabulary-learning strategy.

201. Strong review conclusion

The evidence supports a comparatively well-established short-term vocabulary benefit from retrieval practice, particularly among novice learners. Confidence decreases as follow-up length and learner proficiency increase because delayed studies are fewer, more heterogeneous and underrepresent advanced learners. The review therefore supports a strong short-term claim but not a universal or durable effect across proficiency levels. The highest-value next studies are adequately powered delayed tests in advanced learners using comparable retention outcomes.

202. Systematic-review conclusion job | Evidence density matters

The total number of studies should not create false confidence about a thinner subgroup of the evidence.

203. Worked case F — Meta-analysis with heterogeneity | Meta-analysis

Fictional teaching case.

Forty studies produce a positive pooled effect, but heterogeneity is substantial and larger benefits appear in low-baseline populations.

204. Strong meta-analytic conclusion

The pooled estimate favours the intervention, but substantial between-study variation indicates that the average effect does not describe all populations equally well. Larger effects among lower-baseline groups suggest a possible proficiency boundary, although the moderator evidence remains exploratory because subgroup definitions varied across studies. The practical conclusion is therefore not that the intervention “works everywhere”, but that its average benefit appears context-sensitive and should be tested under more standardised moderator definitions.

205. Worked case G — AI-assisted writing trial | AI 辅助写作

Fictional teaching case.

180 university writers are randomised to AI assistance or a standard word processor. Assisted essays score higher and are completed faster. Four weeks later, unaided writing differs little. AI users also accept a measurable proportion of unsupported generated claims.

206. Weak AI conclusion

AI improves student writing and productivity.

207. Strong AI conclusion

AI assistance improved immediate supported essay performance and reduced completion time, but the four-week unaided task did not establish a durable independent-writing benefit. The factual-error acceptance rate further shows that higher assisted output quality can coexist with reliability costs. The study therefore supports a tool-performance benefit more strongly than a learner-development claim. Future work should separate planning, revision and text-generation components while measuring delayed unaided performance and factual reliability.

208. AI conclusion job | Tool output ≠ user capability

The conclusion protects the distinction between supported performance and independent learning.

209. Worked case H — Computational benchmark paper | Benchmark research

Fictional teaching case.

Model A beats Model B by 0.9 percentage points on Benchmark X, performs similarly on Benchmark Y and degrades more sharply under distribution shift.

210. Weak computational conclusion

Model A is superior to Model B.

211. Strong computational conclusion

Model A achieved a small advantage on Benchmark X but showed no clear advantage on Benchmark Y and degraded more strongly under the tested distribution shift. The results therefore support a benchmark-specific performance gain rather than general model superiority. The principal contribution is the identification of a performance–robustness trade-off that should be evaluated across additional shifts before deployment claims are made.

212. Computational conclusion job | Multi-metric claims need multi-metric closure

“Better” is not a single property when systems trade off accuracy, latency, energy, robustness and reliability.

213. Worked case I — Engineering validation | 工程验证

Fictional teaching case.

A new control system increases throughput by 12% but uses 18% more energy and has not been tested above 35°C.

214. Strong engineering conclusion

The control system increased throughput under the tested operating conditions, but the gain was accompanied by higher energy use and has not been validated at elevated temperatures. Whether the design represents an overall improvement therefore depends on deployment priorities and thermal requirements rather than throughput alone. Future validation should test high-temperature reliability and quantify the throughput–energy trade-off under realistic operating loads.

215. Engineering conclusion job | Trade-off and operating envelope

Engineering conclusions often need to state the operating boundary as part of the claim itself.

216. Worked case J — Diagnostic test | 诊断工具

Fictional teaching case.

A test has high sensitivity and modest specificity in one hospital population.

217. Strong diagnostic conclusion

The test identified most positive cases in the studied population but generated a substantial number of false positives. Its clinical value therefore depends on whether missed cases or unnecessary follow-up carries the greater cost in the intended setting. External validation across populations with different disease prevalence is needed before the observed operating characteristics are generalised.

218. Diagnostic conclusion job | Accuracy is not one number

Sensitivity and specificity create different error costs; the conclusion should not collapse them into “the test is accurate”.

219. Worked case K — Non-inferiority trial | 非劣效试验

Fictional teaching case.

The new treatment meets the prespecified non-inferiority margin and is cheaper to deliver.

220. Strong non-inferiority conclusion

The new treatment met the prespecified non-inferiority criterion relative to the standard treatment and reduced delivery cost under the study conditions. This supports the new treatment as a potentially viable alternative where the chosen non-inferiority margin is clinically acceptable. The finding does not establish that the two treatments are identical, and longer safety follow-up remains necessary before routine substitution.

221. Worked case L — Null primary outcome, positive secondary outcome | 主结果 null

Fictional teaching case.

The primary objective performance outcome is null. Self-reported confidence improves.

222. Weak spin conclusion

The intervention improved learner confidence and shows promise for improving performance.

223. Strong conclusion

The intervention did not produce a clear improvement in the primary objective performance outcome, although participants reported greater confidence as a secondary outcome. The study therefore supports a change in self-perception more strongly than a change in demonstrated capability. Future work should test whether the confidence gain predicts later behaviour or performance rather than treating it as evidence of competence by itself.

224. Worked case M — Failed replication | 未复现

Fictional teaching case.

A replication estimates an effect near zero after an earlier study reported a large effect.

225. Strong replication conclusion

The replication did not reproduce the large original effect under the tested conditions, reducing confidence that the initial estimate generalises broadly. The discrepancy does not by itself establish that the original result was false: sampling variation, implementation differences and population differences remain possible contributors. The combined evidence therefore shifts the field from a large-effect claim toward uncertainty about effect magnitude and boundary conditions.

226. Worked case N — Corpus study | 语料库研究

Fictional teaching case.

Phrase X is more frequent in expert writing than student writing across one disciplinary corpus.

227. Strong corpus conclusion

Phrase X was more frequent in the expert corpus than in the student corpus, identifying a stable distributional difference within the sampled discipline. Frequency alone does not establish that the phrase is intrinsically superior or universally appropriate. The contribution is therefore descriptive and pedagogical: the pattern identifies a candidate feature for closer functional analysis rather than a rule that learners should imitate automatically.

228. Worked case O — Humanities interpretive article | 人文学术文章

Fictional teaching case.

An article argues that a set of late nineteenth-century travel diaries constructed “distance” through recurring metaphors of scale and delay.

229. Strong interpretive conclusion

The diaries do not merely describe geographical distance; they repeatedly construct distance as a temporal and perceptual problem through metaphors of delay, scale and interrupted visibility. Reading these motifs together shifts the interpretation from travel as movement across space to travel as a negotiated experience of incomplete access. This claim is bounded by the selected diary corpus and does not establish the same pattern across travel writing more generally, but it offers a testable interpretive lens for comparison with adjacent archives.

230. Humanities conclusion job | Argument closure, not fake empiricism

Interpretive conclusions still need evidence, scope and contribution, but they should not imitate experimental language where it does not fit.

231. Worked case P — Professional programme evaluation | 专业评估报告

Fictional teaching case.

A tutoring programme improves completion rates modestly, increases tutor workload substantially and has high parent satisfaction.

232. Weak professional conclusion

The programme was successful and should be expanded.

233. Strong professional conclusion

The programme improved completion rates and was rated favourably by parents, but the gain required substantially more tutor time than standard delivery. The evaluation therefore supports the programme’s feasibility and acceptability more strongly than its cost-effectiveness. Expansion should depend on whether the completion benefit justifies the staffing burden and whether similar outcomes can be maintained under a lower-intensity delivery model.

234. Professional conclusion job | Evidence + decision threshold

Evaluation conclusions often need to separate success on one metric from an overall decision.

235. Worked case Q — Policy pilot | 政策试点

Fictional teaching case.

A pilot produces a modest benefit, uncertain distributional effects and high implementation cost.

236. Strong policy conclusion

The pilot produced a modest average benefit, but the implementation burden was high and subgroup estimates were too imprecise to determine whether gains were distributed equitably. The evidence therefore supports continued targeted evaluation rather than immediate scale-up. A larger pilot with prespecified equity outcomes and cost tracking is needed before a system-level recommendation can be made.

237. Worked case R — Measurement study | 测量研究

Fictional teaching case.

A self-report engagement scale correlates moderately with attendance but weakly with observed on-task behaviour.

238. Strong measurement conclusion

The engagement scale captured a component related to attendance but showed limited correspondence with observed on-task behaviour. The measure should therefore not be treated as a complete proxy for behavioural engagement. Its most defensible use is as a self-report perspective that complements rather than replaces direct behavioural evidence.

239. Worked case S — Feasibility study | 可行性研究

Fictional teaching case.

Recruitment, retention and delivery targets are met, but the study is not powered for effectiveness.

240. Strong feasibility conclusion | Do not conclude efficacy from feasibility

The study met its prespecified recruitment, retention and delivery targets, establishing that a larger effectiveness trial is operationally feasible under the tested conditions. Because the study was not designed or powered to estimate effectiveness, outcome differences should remain exploratory. The next step is therefore a fully powered comparative trial rather than routine implementation.

---

Part IV shows that the same closure logic travels across research genres: answer the question actually tested, name the contribution, preserve the strongest boundary, and convert the remaining uncertainty into the next useful decision or study.

Part V — Conclusion failure modes, reviewer tests and revision surgery | 第五部分：Conclusion 最常见的失败模式与修复

241. Most weak conclusions fail after the evidence, not before it | 研究可能是对的，结尾仍然会坏

A paper can have sound Methods, accurate Results and a careful Discussion, then lose credibility in the final paragraph by enlarging the claim, erasing uncertainty or making a recommendation that the study never earned. Conclusion revision therefore deserves its own quality gate.

242. Failure mode 1 — The summary dump | 把全文再说一遍

Symptoms:

• methods repeated;
• every result replayed;
• citations re-listed;
• no higher-level contribution.

Repair:

Delete information the reader already knows unless it performs a new closure job. Preserve only the primary answer, contribution, boundary and next implication.

243. Failure mode 2 — The final certainty jump | 最后突然更肯定

Discussion:

The findings are consistent with a possible short-term benefit.

Conclusion:

The intervention is effective.

This is a certainty mismatch.

244. Certainty audit | Search the final paragraph for upgraded verbs

Compare:

• suggests → shows;
• associated with → improves;
• may contribute → causes;
• supports → proves;
• short-term → durable.

Every upgrade requires new evidence. The Conclusion contains no new evidence, so unexplained upgrades are usually errors.

245. Failure mode 3 — The scope jump | One sample becomes everyone

Discussion:

advanced adult volunteers in one programme.

Conclusion:

students.

Repair by restoring the population or explicitly marking broader generalisation as untested.

246. Failure mode 4 — The time jump | Four weeks becomes “long-term”

Time horizons must survive compression.

247. Failure mode 5 — The construct jump | Score becomes learning

Other common construct inflations:

• confidence → competence;
• attendance → engagement;
• benchmark accuracy → general intelligence;
• clicks → successful use;
• satisfaction → effectiveness.

248. Failure mode 6 — The causal jump | Association becomes cause

Especially common in final sentences written for impact.

249. Failure mode 7 — The policy jump | One effect becomes “should adopt”

A recommendation needs decision evidence beyond the effect estimate.

250. Failure mode 8 — The significance fog | “Important implications” without content

Repair by naming the changed proposition or decision.

251. Failure mode 9 — The novelty boast | “First”, “groundbreaking”, “novel”

Priority claims can be hard to establish and are often unnecessary. Specific contribution is stronger than generic novelty language.

252. Failure mode 10 — The future-work graveyard | A list of unrelated next studies

A strong conclusion selects the next uncertainty that most constrains the current claim.

253. Failure mode 11 — The limitation eraser | “Despite these limitations…”

Weak:

Despite these limitations, the study demonstrates that the intervention is effective.

If the limitations matter, they must change the final claim.

254. Failure mode 12 — The generic “more research” exit | No knowledge frontier

Specify the next test.

255. Failure mode 13 — The new-idea ending | A major theory arrives in the last paragraph

If the idea matters enough to shape the conclusion, it usually belonged in the Discussion where readers could evaluate it.

256. Failure mode 14 — The new-data ending | Never report an unshown result in Conclusion

Conclusion is not a hiding place for analyses that missed Results.

257. Failure mode 15 — The emotional ending | “Exciting”, “encouraging”, “promising” without criteria

These can be acceptable in limited contexts, but they should not replace measurable reasons.

258. Failure mode 16 — The universal lesson | “This shows the importance of…”

Ask whether the study actually tested that broad principle.

259. Failure mode 17 — The slogan ending | Memorable but inaccurate

A slogan may compress away qualifiers that matter. Accuracy wins.

260. Failure mode 18 — The defensive ending | Conclusion argues with imagined reviewers

The final paragraph should state the evidence state, not litigate every possible objection.

261. Failure mode 19 — The exhausted ending | Writer repeats because they have no final abstraction

If you cannot state the paper’s final contribution in one sentence, return to the Discussion and identify what genuinely changed.

262. Failure mode 20 — The over-hedged ending | Everything becomes maybe

Strong evidence deserves strong language. Conclusion should calibrate, not weaken indiscriminately.

263. The “claim ledger” revision method | Build the final paragraph from existing claims

Create four columns:

Claim	Evidence	Boundary	Final wording
near-term effect	randomised one-week result	one programme	improved one-week outcome in this sample
mechanism	indirect participant reports	not isolated	plausible but unresolved
durability	12-week imprecise estimate	uncertain	not established
implementation	high-support setting	routine transport unknown	requires routine-setting test

264. Build the Conclusion only from ledger-approved claims | No orphan claims

If a sentence has no evidence parent in the paper, remove or relocate it.

265. The sentence-parent test | 每句话的 parent 在哪里？

For every Conclusion sentence ask:

Which Results/Discussion claim is its parent?

If you cannot identify one, the sentence may be new, inflated or decorative.

266. The certainty inheritance test | Match parent certainty

A child conclusion sentence cannot be more certain than its parent Discussion claim without justification.

267. The scope inheritance test | Match parent scope

Population, time, outcome and setting should not expand during inheritance.

268. The consequence inheritance test | Limitations must reach the final claim

If a limitation changes mechanism or generalisation in Discussion, the Conclusion should retain that change.

269. The primary-outcome gate | Does the Conclusion foreground the primary answer?

A secondary positive outcome should not displace a null primary outcome at the end.

270. The inconvenient-result gate | Could the conclusion survive if the least convenient result is placed first?

If not, the ending may be narrative spin.

271. The counterevidence gate | Does the final claim survive contradiction?

Conclusion should reflect the full evidence set, not only supportive findings.

272. The rival-explanation gate | Does the conclusion imply mechanism certainty that Discussion rejected?

Mechanism uncertainty must remain visible where it materially affects meaning.

273. The external-validity gate | Does the conclusion travel farther than the sample?

Highlight every broad noun and test it.

274. The recommendation gate | Is “should” earned?

Mark every:

• should;
• must;
• ought;
• needs to be adopted;
• should be implemented.

Then list the decision criteria supporting each.

275. The novelty gate | Is “first” defensible?

If literature search cannot establish priority, state the concrete contribution instead.

276. The importance gate | Can “important” be replaced by a changed proposition?

Usually yes.

277. The final-sentence gate | Does the last sentence close the current paper?

A good last sentence does one of four jobs:

• states contribution;
• states boundary;
• states next test;
• states proportionate decision implication.

278. The stop gate | Would another sentence add a new job?

If not, stop.

279. Reviewer simulation 1 — “The conclusion overstates causality”

Original:

Remote work increased productivity.

Study:

Observational, non-random adoption.

Repair:

Remote-work frequency was associated with higher productivity, but selection into remote work prevents a firm causal interpretation.

280. Reviewer simulation 2 — “The conclusion generalises beyond the sample”

Original:

Structured feedback benefits bilingual learners.

Repair:

Structured feedback improved one-week independent revision among the advanced bilingual adults studied; effects in other learner groups remain uncertain.

281. Reviewer simulation 3 — “Your null result disappears”

Original:

The intervention improved confidence and shows promise.

Primary performance outcome null.

Repair:

The intervention did not improve the primary objective outcome, although confidence increased as a secondary self-report measure.

282. Reviewer simulation 4 — “You claim equivalence from non-significance”

Repair:

The study did not detect a clear difference, but the interval remains too wide to establish equivalence.

283. Reviewer simulation 5 — “Your limitation paragraph changes nothing”

Original limitation:

The study was conducted at one site.

Original conclusion:

The method is effective for advanced learners.

Repair:

The one-site trial supports a near-term effect under high-support conditions; transportability to lower-support settings remains uncertain.

284. Reviewer simulation 6 — “Your conclusion repeats Results”

Delete repeated numbers. Add the contribution and boundary.

285. Reviewer simulation 7 — “Your final recommendation is not supported”

Original:

Hospitals should adopt the test.

Repair:

The test warrants external validation and decision-impact evaluation before routine adoption.

286. Reviewer simulation 8 — “Your future work is generic”

Replace:

Further research is required.

with:

A six-month external-site follow-up is needed to determine whether the observed benefit persists and transports beyond the original programme.

287. Reviewer simulation 9 — “Your conclusion uses a construct you did not measure”

Original:

The training improved resilience.

Measured:

self-reported coping confidence.

Repair:

The training increased self-reported coping confidence; broader resilience was not directly measured.

288. Reviewer simulation 10 — “The conclusion hides harms”

Repair by representing benefit and harm together.

289. Revision surgery: Level 1 — Delete ritual language | 去掉仪式性结尾

Delete or test phrases such as:

• in conclusion;
• to sum up;
• overall;
• it can be concluded that;
• it is clear that.

Keep them only when they perform useful signalling.

290. Revision surgery: Level 2 — Replace praise with contribution | Praise → proposition

important → what changed?

novel → what is new?

promising → which evidence makes further testing worthwhile?

291. Revision surgery: Level 3 — Restore scope words | Add missing qualifiers

Add:

• in this sample;
• under the tested conditions;
• over one week;
• for the measured outcome;
• relative to the comparator.

when needed.

292. Revision surgery: Level 4 — Separate effect from mechanism | Two sentences if necessary

The effect was clear. Its mechanism remains uncertain.

Simple can be powerful.

293. Revision surgery: Level 5 — Make the boundary consequential | Limit → claim change

Not:

The study has limitations.

But:

The short follow-up prevents a durable-learning conclusion.

294. Revision surgery: Level 6 — Convert generic future work into a discriminating test

Ask what uncertainty most changes the current conclusion.

295. Revision surgery: Level 7 — Align abstract, title and conclusion | Three-surface consistency

Read only these three parts. Do they tell the same evidence story?

296. Revision surgery: Level 8 — Back-translate to Chinese | Detect certainty drift

Translate the final English claim back into Chinese. If your internal Chinese becomes stronger than the English or vice versa, inspect the mismatch.

297. Revision surgery: Level 9 — Read aloud | Closure rhythm reveals repetition

A conclusion that repeats the paper often sounds heavy when read aloud. A conclusion that closes cleanly usually has decreasing informational breadth and increasing conceptual clarity.

298. Revision surgery: Level 10 — Remove the final extra sentence | The one-sentence-too-many test

Many conclusions improve when the last generic sentence is deleted.

299. The final reviewer question | “What exactly can I cite this paper for?”

Your conclusion should make that answer clear.

300. Part V operating rule | Final revision principle

Revision is complete when every sentence in the Conclusion has a traceable evidence parent, a controlled scope and a distinct closure job—and when deleting the final sentence would remove real meaning rather than ritual formality.

Conclusion 修订完成的标准是：每一句都有 evidence parent、scope 被控制、closure job 明确；而最后一句如果删掉，会真正丢失 meaning，而不是只少了一个“正式结尾”。

---

Part V turns reviewer criticism into a repair system. Part VI will convert the entire lesson into drills, audits, a mastery rubric, reference floor and final independent-transfer benchmark.

Part VI — Independent transfer, drills, mastery rubric and final benchmark | 第六部分：把 Conclusion 变成可迁移能力

301. The goal is independent closure | 不是背模板

A learner has mastered research conclusions when an unfamiliar paper can be closed accurately without relying on one memorised paragraph frame. The writer should be able to identify the primary answer, compress the contribution, preserve the decisive limitation, scale implication to evidence and stop before the paper becomes larger than itself.

302. The conclusion matrix | 写之前先做四格矩阵

Question	Answer to extract
What did the study establish?	strongest defensible primary answer
What changed in knowledge?	specific contribution
Where does the claim stop?	most consequential boundary
What follows next?	implication, decision or discriminating test

If a proposed conclusion sentence fits none of these jobs, ask why it is there.

303. The evidence-budget worksheet | Final claim budget

Before drafting, write one line for each dimension:

• Causality: experimental / quasi-experimental / observational / descriptive / interpretive.
• Population: exactly who was studied.
• Setting: where and under what support conditions.
• Time: immediate / delayed interval / longitudinal horizon.
• Outcome: exact measured construct.
• Precision: clear / uncertain magnitude / null-compatible / heterogeneous.
• Mechanism: established / supported / plausible / unresolved.
• Decision evidence: benefit / harm / cost / feasibility / reversibility.

Your Conclusion cannot spend beyond this budget.

304. Drill 1 — Answer-first compression | 主问题压缩

Research question:

Does structured feedback improve independent revision one week after support is removed?

Data:

Randomised comparison; clear positive one-week estimate.

Write one sentence that answers the question without mechanism or policy language.

305. Model answer 1

Structured feedback improved one-week independent revision relative to standard feedback in the studied sample.

306. Drill 2 — Add the contribution | 不重复 result

Prior studies measured only supported tasks. Add one contribution sentence.

307. Model answer 2

The study extends earlier work by showing that part of the performance advantage remains when learners revise an unseen task without direct feedback.

308. Drill 3 — Add the boundary | 最大 uncertainty

12-week estimate is smaller and imprecise. Add one boundary sentence.

309. Model answer 3

The smaller and imprecise 12-week estimate leaves longer-term durability unresolved.

310. Drill 4 — Add the next test | 不写“more research”

Planning time also differed between groups. Write a future-study sentence.

311. Model answer 4

A delayed component-controlled trial that matches planning time is needed to distinguish prompt structure from additional cognitive effort.

312. Combine drills 1–4 | Complete four-sentence conclusion

Structured feedback improved one-week independent revision relative to standard feedback in the studied sample. The study extends earlier supported-task work by showing that part of the advantage survives after direct feedback is removed. The smaller and imprecise 12-week estimate leaves durability unresolved, while unequal planning demands prevent clean mechanism attribution. A delayed component-controlled trial that matches planning time is therefore the next critical test.

313. Drill 5 — Observational causality repair | Association → correct closure

Chinese note:

结果表明远程办公提高了生产力，因此企业应该采用远程办公。

Evidence:

Employees self-select remote work; autonomy is a potential confounder.

Reconstruct the Conclusion in English.

314. Model answer 5

Remote-work frequency was associated with higher productivity, but self-selection and baseline autonomy prevent a firm causal interpretation. The findings support stronger causal evaluation of remote-work arrangements rather than a universal adoption recommendation.

315. Drill 6 — Null-result repair | “No difference”

Chinese note:

两种方法没有显著差异，因此效果基本一样。

Interval is wide enough to include meaningful effects in both directions.

316. Model answer 6

The study did not detect a clear between-method difference, but the estimate remains too imprecise to establish equivalence.

317. Drill 7 — Proxy repair | Confidence ≠ competence

Training raises confidence but objective performance does not change clearly.

318. Model answer 7

The training increased self-reported confidence without a corresponding clear improvement in objective performance, supporting a change in self-perception more strongly than a change in demonstrated competence.

319. Drill 8 — Short follow-up repair | “长期有效”

Follow-up ends at four weeks.

320. Model answer 8

The benefit remained detectable at four weeks; persistence beyond that interval was not assessed.

321. Drill 9 — Generalisation repair | “适合所有学生”

Participants are advanced adult volunteers in one high-support programme.

322. Model answer 9

The observed benefit applies directly to the advanced adult volunteers studied; effects in younger, beginner and lower-support populations remain uncertain.

323. Drill 10 — Recommendation repair | “值得广泛推广”

Single efficacy trial; low cost; no long-term data.

324. Model answer 10

The low-cost approach warrants broader routine-setting trials, but current evidence is insufficient for universal adoption because long-term effectiveness remains untested.

325. Drill 11 — Mixed-outcome compression | One positive, one negative

New interface increases click rate but reduces task completion.

326. Model answer 11

The interface increased click activity but reduced successful task completion, showing that the two outcomes moved in opposite directions and should not be compressed into a single claim of improved engagement.

327. Drill 12 — Benefit–harm compression

Treatment improves target outcome by a modest amount and increases adverse events.

328. Model answer 12

The treatment improved the target outcome but also increased adverse events, so its overall value depends on the relative importance and reversibility of these competing effects.

329. Drill 13 — Benchmark scope repair

Model A improves one benchmark by 1% but degrades under distribution shift.

330. Model answer 13

Model A achieved a small benchmark-specific advantage but showed weaker robustness under distribution shift, supporting a performance–robustness trade-off rather than general model superiority.

331. Drill 14 — Qualitative scope repair

Twenty interviewees describe action clarity as useful; five describe overload.

332. Model answer 14

Participants generally experienced action clarity as useful, while accounts of overload indicate that the same structure can become constraining when revision demands exceed learners’ prioritisation capacity.

333. Drill 15 — Replication repair

Original study reports large effect; replication estimates near zero.

334. Model answer 15

The replication did not reproduce the original large effect under the tested conditions, reducing confidence in broad generalisation while leaving sampling, population and implementation differences as possible explanations for the discrepancy.

335. Drill 16 — Feasibility vs effectiveness

Pilot meets recruitment and retention targets; it is not powered for outcomes.

336. Model answer 16

The pilot establishes operational feasibility under the tested conditions but does not establish effectiveness; a fully powered comparative study is the appropriate next step.

337. Drill 17 — “具有重要意义” reconstruction

Chinese note:

本研究具有重要理论与实践意义。

Actual contribution:

The study separates assisted output from delayed independent performance.

338. Model answer 17

By separating assisted output from delayed independent performance, the study narrows what can legitimately be treated as evidence of learner development.

339. Drill 18 — “填补空白” reconstruction

Prior work lacks external validation.

340. Model answer 18

The study adds an external-site validation test that earlier single-site evaluations did not provide.

341. Drill 19 — “进一步验证” reconstruction

Need independent replication, not same-data sensitivity analysis.

342. Model answer 19

Independent replication is needed to determine whether the effect survives a new sample, research team and implementation context.

343. Drill 20 — “相互印证” reconstruction

Survey and interviews agree but share same intervention setting.

344. Model answer 20

Survey and interview findings converged on a similar pattern, although their shared intervention context means common response influences cannot be excluded.

345. The 20-minute Conclusion drill | 20 分钟训练

Time	Task
3 min	write research question and one-sentence earned answer
4 min	identify exact contribution relative to prior gap
4 min	write claim-scope vector
4 min	select one consequential boundary
3 min	write one next implication/test
2 min	delete ritual final sentence

346. The 45-minute growth session | 45 分钟增长模式

Time	Task
7 min	reconstruct evidence budget
8 min	build claim ledger
8 min	draft four-sentence conclusion
7 min	run causality/scope/construct audits
7 min	run Mandarin→English reconstruction audit
5 min	align title, abstract and conclusion
3 min	read aloud and apply stop gate

347. The 90-minute deep session | 90 分钟深度模式

Time	Task
10 min	map question → Results → Discussion final state
10 min	build evidence-budget vector
10 min	identify contribution and changed proposition
10 min	rank limitations by consequence
10 min	draft two alternative conclusion architectures
10 min	compare recommendation thresholds
10 min	bilingual back-translation and reconstruction
10 min	reviewer simulation
10 min	final compression and stop gate

348. Seven-day conclusion cycle | 七天循环

1. Day 1 — Answer: write one-sentence conclusions from five Results sections.
2. Day 2 — Contribution: turn “important / gap / novel” into specific changed propositions.
3. Day 3 — Scope: audit population, setting, time, outcome and comparator.
4. Day 4 — Certainty: repair causal, null and equivalence errors.
5. Day 5 — Implication: separate implication from recommendation and decision threshold.
6. Day 6 — Bilingual: reconstruct ten Chinese closing phrases from evidence states.
7. Day 7 — Transfer: write an unseen full Conclusion and compare with the published version.

349. Twelve-week C1–C2 conclusion progression | 12 周进阶

Weeks	Focus	Output
1–2	answer vs summary	50–100 word closures
3–4	contribution specificity	gap-to-contribution rewrites
5–6	causality, scope, null/equivalence	claim-control portfolio
7–8	implications, recommendations, future tests	decision-aware conclusions
9–10	Mandarin–English reconstruction	bilingual conclusion portfolio
11–12	genre transfer and peer review	submission-ready conclusions across three disciplines

350. Monthly benchmark | 每月 benchmark

Select an unfamiliar research paper. Read its Introduction, Methods, Results and Discussion but hide the published Conclusion.

1. Write the research question in one sentence.
2. State the primary result without interpretation.
3. State the final Discussion interpretation.
4. Build the claim-scope vector.
5. Identify the paper’s exact contribution.
6. Rank the three most important limitations.
7. Select the one limitation that most changes the final claim.
8. Write a four-sentence Conclusion.
9. Compress it to two sentences.
10. Reconstruct it in Simplified Chinese.
11. Rebuild it in English without looking at the first draft.
12. Compare certainty and scope across the two English versions.
13. Read the paper’s actual Conclusion.
14. Mark where the authors are stronger or weaker than your version.
15. Decide which version better matches the evidence and why.

351. The one-page Conclusion operating system | 一页操作系统

1. Recover the primary question.
2. Write the strongest defensible answer.
3. Name the contribution as a changed proposition.
4. Build the scope vector.
5. Preserve null, mixed or adverse findings that materially change meaning.
6. Separate effect from mechanism.
7. Choose the most consequential boundary.
8. Separate implication from recommendation.
9. If recommending action, state the decision logic.
10. Turn the largest uncertainty into a discriminating next test.
11. Do not introduce new data or a major new theory.
12. Audit causality, population, time, outcome and construct.
13. Audit Mandarin→English certainty drift.
14. Align title, abstract and conclusion.
15. Delete generic final sentences.
16. Stop when every remaining sentence has a distinct closure job.

352. The conclusion mastery rubric | 100-point rubric

Dimension	Points	Mastery evidence
Primary-question fidelity	15	final answer directly resolves the actual research question
Evidence alignment	15	claim strength matches design, estimate and uncertainty
Contribution specificity	10	states exactly what knowledge changed
Scope control	15	population, setting, time, outcome and comparator preserved
Counterevidence honesty	10	null, mixed or adverse findings remain visible when consequential
Boundary quality	10	largest limitation changes the final claim
Implication/recommendation control	10	decision language is proportionate and justified
Future-work specificity	5	next study resolves the largest uncertainty
Bilingual epistemic precision	5	no literal Chinese phrase inflates certainty or significance
Closure and stop control	5	no repetition, new topic or ritual extra sentence

353. What 90–100 looks like | C2-ready conclusion

The conclusion can be read independently as an accurate final state of the paper. A sceptical expert can trace every claim to evidence, see exactly where the claim stops and understand what the study contributes without accepting every preferred interpretation. The last sentence adds real closure rather than ceremony.

354. What 75–89 looks like | Strong C1 conclusion

The main answer is accurate and contribution clear, but one local weakness remains: perhaps scope is slightly broad, the limitation does not fully affect the conclusion, future work is generic or one Mandarin-derived phrase inflates significance. Repair is local rather than architectural.

355. What 60–74 looks like | Fluent but unstable ending

The conclusion sounds academic but relies on stock phrases such as important implications, more research is needed, clearly shows or should be widely adopted. The paper’s evidence is being compressed into rhetoric rather than controlled propositions.

356. Below 60 | Summary or advocacy, not research closure

The ending either repeats the paper without contribution or turns the study into a larger claim than the evidence supports. Return to the claim ledger and rebuild from evidence parents.

357. Self-marking protocol | Evidence before points

To award points, quote the exact sentence that performs each job:

• primary answer;
• contribution;
• boundary;
• implication;
• next test.

If you cannot point to a sentence, do not award the points.

358. Compression challenge | 400 words → 150 → 75

Write a 400-word conclusion. Compress it to 150 words while preserving answer, contribution, boundary and next test. Then compress it to 75 words. Compare:

• Did causality strengthen?
• Did scope widen?
• Did a limitation disappear?
• Did “more research” replace a specific next study?

The exercise trains compression without distortion.

359. Expansion challenge | 75 words → 300 without repetition

Start with a calibrated 75-word conclusion. Expand by adding only new closure jobs:

• more precise contribution;
• one consequential boundary;
• one decision implication;
• one specific next test.

Do not add methods recap or Results replay.

360. The adversarial conclusion test | Make it too strong, then identify assumptions

Valid:

The intervention improved one-week revision in this sample.

Overclaim:

The intervention creates lasting independent writers.

Extra assumptions introduced:

• one-week → lasting;
• revision score → whole-writer capability;
• sample → learners generally;
• effect → enduring mechanism.

361. The reverse adversarial test | Restore earned confidence

Over-hedged:

The intervention may perhaps have had some possible influence on the measured outcome.

If the randomised estimate is clear:

The intervention improved the measured one-week outcome.

362. The “one sentence too many” exercise | 最后一句删除测试

Take your Conclusion and remove the final sentence. If nothing meaningful disappears, leave it deleted. If a specific contribution, boundary or next test disappears, restore it.

363. The back-translation exercise | English → Chinese → English

Translate your final English conclusion into natural Simplified Chinese. Hide the original English and reconstruct the Conclusion again from the Chinese meaning. Then compare the two English versions for:

• certainty;
• causality;
• scope;
• significance;
• recommendation strength.

364. The source-ownership exercise | Who owns each claim?

Mark each sentence as:

• our data establish;
• our interpretation;
• broader literature supports;
• decision implication;
• future hypothesis.

Do not blur these ownership states.

365. The title–abstract–conclusion triangle | Three surfaces, one evidence state

Print or copy only:

1. title;
2. abstract final two sentences;
3. conclusion.

Ask whether they agree on causality, population, time horizon, outcome and mechanism certainty.

366. The introduction–conclusion loop test | Open question vs closed answer

Place the final paragraph beside the final Introduction paragraph. The Conclusion should answer the problem the Introduction made necessary.

367. The citation test | What can another scholar safely cite this paper for?

Complete:

This paper provides evidence that ______, under ______, while ______ remains uncertain.

If that sentence is hard to write, your Conclusion may not yet be precise enough.

368. The decision-maker test | What can someone do now?

Write two lines:

Safe action now: ______.

Action that still needs evidence: ______.

This is especially useful for policy, education, health, engineering and programme evaluation.

369. The theory-update test | Model before → model after

Write:

Before this paper, the field could plausibly say ______.

After this paper, the strongest defensible update is ______.

This converts vague “theoretical significance” into a concrete knowledge change.

370. The uncertainty-update test | Uncertainty should become more specific

Good research does not simply reduce uncertainty. It often transforms broad uncertainty into a narrower, testable uncertainty.

Before:

Does structured feedback work?

After:

Near-term transfer is supported; mechanism and six-month durability remain open.

371. The final assignment | 完整作业

Choose an empirical research paper, a permitted dataset or a fictional study with enough information to reconstruct the evidence state.

1. Write the research question.
2. Identify the primary outcome.
3. Write the strongest defensible answer.
4. Build the evidence-budget worksheet.
5. Build the claim-scope vector.
6. State the contribution as a changed proposition.
7. Identify the most inconvenient result.
8. Identify the most consequential limitation.
9. Explain how that limitation changes the final claim.
10. Separate effect from mechanism.
11. Write one implication.
12. Write one recommendation only if decision evidence supports it.
13. Design one next study that discriminates the largest uncertainty.
14. Write a 250–400 word Conclusion.
15. Compress it to 120 words.
16. Compress it to 70 words.
17. Translate the 120-word version into Simplified Chinese.
18. Reconstruct it in English from the Chinese without seeing the original.
19. Compare certainty and scope.
20. Align title, abstract and conclusion.
21. Apply the stop gate.

372. Final unseen-paper benchmark | No model answer

Select a research paper you have not studied before. Hide the Conclusion. Read the Introduction, Methods, Results and Discussion. Then produce:

• one-sentence primary answer;
• one-sentence contribution;
• one-sentence boundary;
• one-sentence next implication;
• a 120-word final Conclusion.

Only after finishing should you read the published Conclusion. Compare which version more faithfully preserves evidence, scope and uncertainty. The goal is not to imitate the authors. The goal is to become able to judge the closure independently.

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

• University of Manchester Academic Phrasebank | Writing Conclusions — conclusion functions, significance, limitations and recommendations/future work.
• University of Manchester Academic Phrasebank | Being Cautious — calibrated academic claims and uncertainty.
• Purdue Writing Lab | Active Reading as a Graduate Student — conclusions as concise big takeaways for readers who already know the work.
• Nature Scitable | Scientific Papers — conclusion as higher-level interpretation connected to the Introduction’s motivation.
• Nature Physics | Elements of Style — perspective rather than lazy repetition.
• Nature Cancer | The Craft (and Art) of Scientific Writing — findings, implications, caveats, future paths and avoiding hype.
• EQUATOR Network | Reporting Guidelines — design-specific transparency and reporting context.

374. Canonical eduKate research-writing route | 研究写作路线

• Lesson 012 | Literature Review as Scholarly Conversation
• Lesson 013 | Academic Argument That Survives Counterevidence
• Lesson 014 | Calibrate Certainty Without Becoming Vague
• Lesson 015 | Methods, Results and Discussion in Different Jobs
• Lesson 016 | Abstract Compression
• Lesson 017 | Research Introduction
• Lesson 018 | Methods for Reproducibility
• Lesson 019 | Results That Reveal the Pattern
• Lesson 020 | Discussion Without Overclaiming

375. SEO language map | 本课自然覆盖的搜索意图

This lesson naturally serves readers searching for how to write a conclusion, research paper conclusion, conclusion vs discussion, conclusion vs abstract, how to end a research paper, dissertation conclusion, thesis conclusion, research contribution, limitations in conclusion, future research in conclusion, academic conclusion examples, avoid overclaiming, causal language in conclusions, generalisability, practical implications, research recommendations, C1 academic writing, C2 academic English, academic English for Chinese speakers, English for Mandarin speakers, Conclusion 怎么写, 论文结论怎么写, 研究结论, 结论与讨论区别, 结论与摘要区别, 学术英语结论, 中文母语学术英语 and C1 C2 research writing.

376. Final quality gate | 最终质量门

Before submission, answer yes or no:

1. Does the first substantive sentence answer the actual research question?
2. Does the final claim use no stronger causal verb than the design supports?
3. Is the measured outcome still the measured outcome?
4. Are population, setting and time boundaries still visible?
5. Did the primary outcome keep its priority?
6. Did a null, mixed or harmful result disappear?
7. Is the contribution specific rather than praised?
8. Does the key limitation actually reduce or redirect the claim?
9. Is implication separated from recommendation?
10. Does every “should” have decision logic?
11. Is future work a specific test rather than a wish list?
12. Did you avoid major new evidence or theory?
13. Does the Conclusion agree with the Abstract and title?
14. Did literal Chinese closing language inflate certainty?
15. Does the last sentence perform a real closure job?
16. Would the Conclusion become better if the final sentence were deleted?

377. The final principle | A conclusion is where intellectual restraint becomes visible

The best research Conclusion is not the paragraph that makes the study sound largest. It is the paragraph that makes the final state of knowledge easiest to see.

It tells the reader what the evidence now permits, what it still forbids, what changed because the study exists and what the next serious question has become.

最好的研究 Conclusion，不是把研究写得最大，而是把最终 knowledge state 写得最清楚：什么现在可以说，什么仍然不能说，这项研究真正改变了什么，以及下一步最值得解决的问题是什么。

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

You can read an unfamiliar research paper and write its Conclusion without copying the authors, while keeping:

• the primary answer accurate;
• the contribution specific;
• the scope controlled;
• certainty calibrated;
• counterevidence visible;
• recommendations proportionate;
• future work discriminating;
• Mandarin and English epistemic force aligned;
• the final sentence genuinely final.

At that point, conclusion writing is no longer a formula at the end of a paper. It is evidence-controlled closure.

---

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