360 Secure SkillHub
360 安全技能中心

Results Claim & Hedging Checker

专为心理与STEMM学术论文结果段打造,核心诊断五类问题:模糊语缺失/过度、主张与证据不匹配、因果语言不符研究设计、混入理论解释、使用主观评价语。典型场景:提交结果段草稿求审阅,或提及模糊语、过度主张等需求。输入支持中英文本及可选研究设计、期刊信息,输出含评分、问题、建议的标准格式报告。

seulkilu
seulkilu
热度 9.17
最近更新 2026-08-27 19:39:30
分类:效率依赖:读写文件 / 内部调用大模型应用场景:校对润色 / 报告撰写

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add results-claim-hedging-checker
Metadata
name
results-claim-hedging-checker
description
Academic writing revision advisor for the Results section of psychology and STEMM research papers. Checks claim-hedging alignment, causal language appropriateness, statistical reporting completeness, and interpretation boundaries. Triggers when the user asks to review, revise, or check a Results section draft, or mentions hedging, claim strength, overclaiming, or causal language in academic writing.
agent_created
true

Results Claim & Hedging Checker

Purpose

Diagnose and revise the Results section of psychology / STEMM research papers for five recurring problem types: (1) missing or excessive hedging, (2) claim strength mismatched to statistical evidence, (3) causal language inappropriate for the design, (4) theoretical interpretation leaking into Results, and (5) subjective evaluative language. Produce a sentence-level audit with risk ratings and concrete revision suggestions, grounded in 18 curated examples from 8 classic psychology papers.

When to Use

  • User submits a Results section draft (full or partial) and asks for review or revision.
  • User mentions "hedging," "claim strength," "overclaiming," "causal language," or "Results section" in the context of academic writing.
  • User asks to check whether their Results section is "too strong" or "too speculative."
  • User wants to align their writing with conventions observed in published psychology / STEMM papers.

When NOT to Use

  • Reviewing Introduction, Discussion, or Method sections (different conventions apply).
  • General proofreading for grammar, spelling, or formatting (use general editing tools instead).
  • Checking the validity of the statistical analysis itself (this skill evaluates reporting, not analysis).
  • Non-academic writing tasks (e.g., blog posts, reports, marketing copy).

Inputs

  • Required: A Results section draft in English or Chinese (plain text, Markdown, or a file path to .md/.txt/.docx).
  • Optional: The paper's research design (experimental, correlational, cross-sectional, longitudinal) — improves causal-language diagnosis accuracy.
  • Optional: Target journal or style guide (e.g., APA 7th) — adjusts formatting expectations for statistical reporting.

Workflow

Step 1 — Parse and Segment

  1. Read the submitted Results draft.
  2. Split the text into individual sentences or minimal clause-level units. Each unit becomes one diagnostic record.
  3. Tag each unit with its sequential position (e.g., S1, S2, S3 …) for traceability.

Step 2 — Classify Each Unit

For every sentence, check against the five diagnostic dimensions. A sentence may trigger more than one dimension.

Dimension What to detect Reference examples
D1. Hedging Missing hedging where indirect evidence is used; or excessive hedging that undermines clear findings F-01, F-05, F-07, F-09, F-16
D2. Claim Strength vs. Evidence Overclaims beyond what the statistics support; strong claim words ("strong evidence," "robust finding") used without corresponding evidence; causal leaps from correlational data F-03, F-10, F-15
D3. Causal Language Strong causal verbs ("caused," "produced," "leads to," "determines") used when design does not support causal inference, or even when it does but Results convention favors neutral phrasing F-06, F-11, F-13
D4. Interpretation in Results Theoretical mechanisms, "because" explanations, hypothesis restatements, or discussion-level conclusions appearing in Results F-04, F-08, F-12, F-18
D5. Subjective / Evaluative Language Evaluative adjectives ("instructive," "conspicuous," "remarkable"), subjective hedges ("It is helpful to consider," "Interestingly,"), or absolute expressions ("leave no room for doubt," "prove") F-02, F-14, F-15, F-17

Step 3 — Score Each Unit

Apply the scoring rubric (below) to assign a 1—5 分 per dimension per sentence(5 分为最高,1 分为最低)。如果某个维度没有问题,该维度应得 5 分。不得在输出中使用"2 分 / Pass"这种含糊表述——直接给出 1—5 的数字分数。如果某句在所有五个维度上均为 5 分,则该句无需出现在报告中(除非用户要求全文审计)。

Step 4 — Generate Revision Suggestions

For each flagged sentence:

  1. Quote the original text.
  2. State the diagnosed problem and the dimension it falls under.
  3. Provide a concrete revision suggestion — not just "add hedging" but the actual revised sentence.
  4. If available, cite the matching reference example (by Example ID) and the reusable rule extracted from it.

Step 5 — Compile Output

Assemble all diagnostic results into the output format specified below. 最终输出必须严格使用以下六个标题,不得使用其他自定义标题(Dimension Score / Key Problems / Evidence from Draft / Example-based Comparison / Revision Suggestions / Priority Level),以便汇总 Skill 统一整合各子 Skill 的诊断结果。不得添加额外的自定义标题或省略任何一个标准标题。

Dimension Score 部分必须:

  1. 先给出 D1—D5 每个维度的分数(1—5 分,5 分为最高)。
  2. 再给出一个总体分(1—5 分),综合各维度评分得出。
  3. 附一句话总评。
  4. 如果用户草稿质量较高(各维度均为 5 分或仅有轻微风格问题),总体分应相应较高(4 分或 5 分)。
  5. Summary 中不得出现"2 分 / Pass"的标记——直接使用 1—5 的数字分数。

Scoring Rubric

每个维度采用 1—5 分制,5 分为最高(无问题),1 分为最低(严重问题)。每个维度独立评分后,综合各维度给出一个总体分(1—5 分),并附一句话总评。

分值 含义 行动要求
5 无问题:声称强度与证据匹配,hedging 恰当,因果语言与设计一致,无理论解释渗入,无主观评价词 无需修改
4 轻微:风格偏好层面的微小问题,不影响科学准确性 可选修改;低优先级
3 中度:声称略超出证据支持,或 hedging 缺失/过度 建议修改;说明原因
2 较严重:因果语言与设计不匹配,或相关数据写成因果断言,或 Results 中出现理论解释 推荐修改;解释误导风险
1 严重:绝对化断言(prove, definitely),因果跳跃严重,使用强动词且无对应证据 必须修改;解释误导风险

注意:如果某个维度没有问题,该维度应得 5 分,不得使用"Pass""2 分 / Pass"等含糊表述。

Dimension-Specific Criteria

D1 — Hedging

  • 5: hedging 使用恰当,既不过度也不缺失。
  • 4: hedging 略弱或略强,属风格偏好。
  • 3: 间接证据(中介、机制、观察数据)处缺少 hedging。
  • 2: 相关数据上直接因果断言且零 hedging。
  • 1: 无任何 hedging 的绝对化因果断言。

D2 — Claim Strength vs. Evidence

  • 5: 所有声称都有充分的统计证据支持,声称强度与证据匹配,无过度声称或因果跳跃。
  • 4: 声称强度与证据基本匹配,仅有微小风格问题。
  • 3: 部分声称强度略超出证据支持(如使用 "strong evidence" 但未提供效应量或置信区间,或从相关结果推断因果)。
  • 2: 声称明显超出证据支持(如使用 "robust finding" 但统计证据薄弱,或相关数据写成因果断言)。
  • 1: 大量声称缺乏证据支持,因果跳跃严重,使用 "prove""cause""definitely" 等强动词且无对应证据。

职责边界:本维度不负责统计报告格式的完整性检查(如效应量、置信区间的具体格式)。单纯缺少效应量或置信区间且未使用强声称词时,不作为扣分项。

D3 — Causal Language

  • 5: 因果语言与设计类型完美匹配,或主动声明不可做因果推断。
  • 4: 真实验中使用 "resulted in" 等略强动词,可更中性但非错误。
  • 3: 准实验或观察性数据中使用 "leads to""causes" 等因果词。
  • 2: 横断面相关数据上使用 "caused""produced""determined" 等强因果词。
  • 1: 观察性数据上使用绝对化因果断言且无任何限制声明。

行为用法豁免:当触发词(如 "produced")用于描述被试行为(如 "participants produced responses")而非因果推断时,不视为问题,该维度记 5 分。

D4 — Interpretation in Results

  • 5: 纯数据报告,无理论解释渗入;混淆变量处主动声明限制。
  • 4: 单句中嵌有一个简短的理论从句,但不影响整体报告性质。
  • 3: 完整的 "because X … therefore Y" 解释句出现在 Results 中。
  • 2: 段落级别的理论讨论或假设重述出现在 Results 中。
  • 1: Results 中出现结论性总结 + 理论解释,完全偏离结果报告。

D5 — Subjective / Evaluative Language

  • 5: 用词完全客观,无评价性语言或绝对化表达。
  • 4: 单个评价性形容词(如 "interesting")出现在其他客观的句子中。
  • 3: 主观引导语(如 "It is instructive to …""It is helpful to consider …")。
  • 2: 评价性形容词与主观框架组合使用。
  • 1: 绝对化表达(如 "leave no room for doubt""proved""definitely")+ 主观框架。

Output Format

最终输出必须严格使用以下六个标题,不得使用其他自定义标题,以便汇总 Skill 统一整合各子 Skill 的诊断结果。

# Results Claim & Hedging Audit

## Dimension Score

评分:1—5 分(**5 分为最高,1 分为最低**)。先给出 D1—D5 每个维度的分数,再给出一个总体分,并附一句话总评。不得使用"2 分 / Pass"等含糊表述。

> 示例(高质量草稿,各维度均无问题):
> - D1 Hedging: 5 分 — hedging 使用恰当,无过度或缺失。
> - D2 Claim Strength vs. Evidence: 5 分 — 声称强度与证据匹配,无过度声称。
> - D3 Causal Language: 5 分 — 因果语言与设计类型匹配,无问题。
> - D4 Interpretation: 5 分 — 纯数据报告,无理论解释渗入。
> - D5 Subjective Language: 5 分 — 用词客观,无评价性语言。
> - **总体分: 5 分** — 草稿质量较高,声称强度与证据匹配,无过度声称或因果跳跃。

> 示例(存在问题的草稿):
> - D1 Hedging: 3 分 — 间接证据处缺少 hedging。
> - D2 Claim Strength vs. Evidence: 2 分 — 使用 "strong evidence" 但未提供对应统计证据。
> - D3 Causal Language: 4 分 — 因果语言基本匹配,个别动词可更中性。
> - D4 Interpretation: 5 分 — 无理论解释渗入。
> - D5 Subjective Language: 3 分 — 出现 "instructive" 等评价性形容词。
> - **总体分: 3 分** — 存在多处声称-证据不匹配和 hedging 缺失,建议逐句修改。

## Key Problems

主要问题:逐条列出。

> 示例:
> 1. S4 使用 "strong evidence" 但未提供效应量或置信区间支撑。
> 2. S6 的 "produced" 虽在结构启动文献中属惯用行为动词,但匹配 D3 触发词清单。

## Evidence from Draft

原文证据:摘录相关句子。

> 示例:
> - **S4**: "It showed a strong main effect of prime type (Estimate = .56, SE = .16, z = 3.60, p < .001)."
> - **S6**: "Pair-wise comparisons indicated that participants produced fewer DO responses following PO-An primes."

## Example-based Comparison

与 examples 的对照:引用具体 Example ID 并说明对应关系。

> 示例:
> - **S4 vs. F-07**(Elkin et al., 1989):F-07 使用 "there was evidence of significant superiority" 进行适度 hedging,而 S4 直接使用 "strong",声称强度超出证据支持。
> - **S6 vs. F-01**(Milgram, 1963):F-01 使用 "showed signs of" 描述被试行为,是规范的行为描述;S6 的 "produced" 同样用于描述被试行为,属行为用法豁免。

## Revision Suggestions

修改建议:逐句给出可操作的替换方案。

> 示例:
> - **S4 修改前**: "It showed a strong main effect of prime type (Estimate = .56, SE = .16, z = 3.60, p < .001)."
>   **S4 修改后**: "It showed a main effect of prime type (Estimate = .56, SE = .16, z = 3.60, p < .001)."
> - **S6 修改前**: "participants produced fewer DO responses following PO-An primes"
>   **S6 修改后**: "participants gave fewer DO responses following PO-An primes"

## Priority Level

修改优先级:按高、中、低排列。

> 示例:
> - **高**: S4 — 强声称词 "strong" 缺乏证据支撑,可能误导读者高估效应大小。需删除 "strong" 或补充效应量。
> - **中**: S6 — "produced" 触发 D3 但属行为用法豁免,替换为 "gave" 可消除歧义,非必须。
> - **低**: 全文风格统一性检查,建议后续通读确认所有声称词与证据匹配。

Reference Files

  • references/examples/examples_memberF.md — 18 curated examples (F-01 ~ F-18) from 8 classic psychology papers. Load this file before generating revision suggestions to match user sentences against canonical patterns.
  • references/rubric.md — (Placeholder) Extended scoring rubric with domain-specific rules.
  • references/checklist.md — (Placeholder) Pre-submission checklist for authors.

Limitations

  1. Domain scope: Examples are drawn from classic psychology papers (social, clinical, developmental, cognitive). Conventions in other STEMM fields (e.g., genomics, neuroscience) may differ; apply judgment.
  2. Language scope: Optimized for English academic writing. Chinese-language Results sections follow related but distinct conventions; use with caution.
  3. Statistical analysis validity: This skill evaluates how results are reported, not whether the analysis is correct. It cannot detect p-hacking, violations of assumptions, or inappropriate tests.
  4. Design inference: If the user does not provide the research design, causal-language diagnosis (D3) defaults to conservative — flagging any strong causal verb. Provide design information for more accurate scoring.
  5. False positives: Sentences that legitimately summarize results at the end of a Results subsection may be flagged for D4 (Interpretation). Use human judgment to distinguish concluding summaries from theoretical discussion.
  6. Example coverage: The 18 examples cover five dimensions but do not exhaust all possible Results-section issues (e.g., table/figure referencing, negative-result reporting, replication language). Expand the example set as needed.
  7. Statistical reporting format is out of scope: 本 Skill 不负责统计报告格式检查,包括效应量、置信区间的具体报告格式和完整性。这些由 results-statistics-convention-checker 负责。
  8. D2 仅在强声称与证据不匹配时触发: 本 Skill 仅在作者使用强声称词(如 "strong effect""clear evidence""robust finding""definitely")却未提供相应统计证据时,从 claim strength 与证据匹配的角度进行提示。如果结果本身未使用强声称,单纯缺少效应量或置信区间不应作为本 Skill 的扣分项。
  9. D3 触发词的行为用法不扣分: 当触发词(如 "produced")用于描述被试行为(如 "participants produced responses")而非因果推断(如 "the prime produced the effect")时,不应视为问题。判定关键在于主语是否为实验操纵变量及其是否暗示因果关系。