check-originality

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Check Originality

检查原创性

Multi-method originality check that catches the overlap which actually gets papers desk-rejected or retracted — before an editor's integrity scan does.

Read this honestly and tell the user: this is not iThenticate or Turnitin. It has no access to any private publisher corpus, so it cannot guarantee a passage is original against all published literature. What it does reliably is detect overlap against sources you can name or fetch, internal text recycling, and distinctive phrases that are publicly searchable. For a journal's official originality report you still need the publisher's tool — this gets your draft clean first.

采用多种方法进行原创性检查，能够检测出那些真正会导致论文被desk-rejected或撤稿的重叠内容——在编辑进行完整性扫描之前就发现问题。

请如实告知用户：本工具并非iThenticate或Turnitin。它无法访问任何私有出版商语料库，因此无法保证某一段落相对于所有已发表文献都是原创的。它能可靠完成的工作是：检测与你指定或获取的来源之间的重叠内容、内部文本复用情况，以及可公开搜索到的独特短语。若要获取期刊官方的原创性报告，你仍需使用出版商提供的工具——本工具的作用是先让你的草稿达到合格标准。

When to use

使用场景

Before submission, to catch accidental copy-paste and close paraphrase.
When reusing your own prior work (text recycling / self-plagiarism is a real violation at most venues even though it's "your" text).
When a co-authored draft may contain a collaborator's verbatim text from elsewhere.
After heavy AI drafting, to confirm nothing was reproduced verbatim from a source.

投稿前，检测意外的复制粘贴和过于接近原文的改写。
当你复用自己之前的研究成果时（文本复用/自我抄袭在大多数场合都是违规行为，即便文本属于你自己）。
合著草稿中可能包含合作者从其他地方直接引用的文本时。
大量使用AI生成草稿后，确认没有从来源中直接复制内容。

The detection methods (run several — that's what "catch it properly" means)

检测方法（需运行多种方法——这才是“准确检测”的含义）

See references/detection-methods.md for the full method and thresholds.

Source overlap — shingle (k-gram) comparison against sources the user provides or that
```
fetch-paper
```
pulls (open-access only). Reports matched passages + overlap %.
Self-plagiarism / recycling — near-duplicate passages against the author's own prior papers, and internal duplication within the draft.
Quote & paraphrase integrity — verbatim spans that lack quotation marks or a citation; paraphrases that stay too close to the source's wording.
Distinctive-phrase external search — pull the draft's most distinctive long phrases and search the web / Semantic Scholar for verbatim matches the local check can't see.
Common-knowledge vs needs-citation — flag factual claims presented without a citation that likely need one (hand off specifics to
```
verify-citations
```
).

查看references/detection-methods.md获取完整方法和阈值。

来源重叠检测——将草稿与用户提供的来源或
```
fetch-paper
```
获取的来源（仅限开放获取资源）进行shingle（k-gram，n元语法分片）比对。报告匹配段落及重叠百分比。
自我抄袭/文本复用检测——检测与作者自己之前论文的近似重复段落，以及草稿内部的重复内容。
引用与改写完整性检测——检测未加引号或未引用的直接引用内容；检测与原文措辞过于接近的改写内容。
独特短语外部搜索——提取草稿中最具辨识度的长短语，并在网络/Semantic Scholar上搜索本地检测无法发现的完全匹配内容。
常识内容与需引用内容区分——标记未加引用但可能需要引用的事实性声明（具体细节交由
```
verify-citations
```
处理）。

Process

流程

Gather comparison sources. Ask the user for: their own prior papers (for self-plagiarism), any suspected sources, and the paper's own cited references (fetch the open-access ones with
```
fetch-paper
```
). More sources = a more meaningful check; say so.
Run source + self overlap:
```
python3 scripts/overlap_check.py --draft paper.tex --against prior1.tex prior2.tex source.txt --self --distinctive 15
```
It prints flagged sources (overlap %), the matched passages, internal near-duplicates, and distinctive phrases. Exit code 2 if anything is flagged.
External pass. Take the emitted distinctive phrases and search them (web + Semantic Scholar via
```
find-papers
```
) for verbatim hits the local pass couldn't see. Record any matches.
Quote/paraphrase review. For each flagged passage, decide: is it a legitimate quote (then it needs quotation marks + citation), an over-close paraphrase (rewrite in the author's own words), or copied (remove/rewrite + cite). Self-overlap with prior work → flag as recycling and advise disclosure or rewrite.
Report to
```
paper-workspace/review/originality-report.md
```
in the user's paper workspace: each finding with source, the overlapping text, severity, and the fix — plus the explicit scope caveat.

收集比对来源。向用户索要：他们自己之前的论文（用于检测自我抄袭）、任何疑似来源，以及论文自身引用的参考文献（通过
```
fetch-paper
```
获取开放获取的文献）。来源越多，检测结果越有意义，请告知用户这一点。
运行来源+自我重叠检测：
```
python3 scripts/overlap_check.py --draft paper.tex --against prior1.tex prior2.tex source.txt --self --distinctive 15
```
该命令会输出标记的来源（重叠百分比）、匹配段落、内部近似重复内容以及独特短语。若有内容被标记，退出码为2。
外部检测环节。使用输出的独特短语进行搜索（通过
```
find-papers
```
在网络和Semantic Scholar上搜索），查找本地检测未发现的完全匹配内容。记录所有匹配结果。
引用/改写审核。针对每个标记段落，判断：它是否是合法引用（此时需要添加引号和引用标注）、是否是过于接近原文的改写（用作者自己的语言重写），或是复制内容（删除/重写并添加引用）。与自身之前工作的重叠内容→标记为文本复用，并建议披露或重写。
生成报告：在用户的论文工作区的
```
paper-workspace/review/originality-report.md
```
中生成报告：每个发现需包含来源、重叠文本、严重程度以及修复建议——同时需明确说明工具的适用范围限制。

Output

输出结果

An originality report listing, per finding: the source, the overlapping passage, overlap %, type (external copy / self-recycle / unquoted / close-paraphrase), severity, and the recommended fix. Always includes the scope/limitations note and a reminder that the publisher's own tool is still authoritative.

一份原创性报告，每个发现需列出：来源、重叠段落、重叠百分比、类型（外部复制/自我复用/未加引号/接近原文的改写）、严重程度以及建议修复方案。报告中始终需包含适用范围/局限性说明，并提醒用户出版商的工具仍是权威标准。

Guardrails

约束规则

Never claim a clean result means the paper is plagiarism-free — only that no overlap was found against the sources checked. State the limitation every time.
Comparison sources are processed transiently; never bundle or commit third-party paper text (see
```
fetch-paper
```
). Only the user's own papers and open-access sources are fetched.
Never rewrite a flagged passage to merely evade detection while keeping the copied idea uncited — the fix is proper quotation/citation or genuine rephrasing, and surfacing it to the author.
Self-overlap with the author's own work is reported as an ethics/recycling issue, not silently passed.

切勿声称检测结果合格就意味着论文完全无抄袭——只能说明在已检查的来源中未发现重叠内容。每次都需说明这一局限性。
比对来源仅做临时处理；切勿打包或提交第三方论文文本（详见
```
fetch-paper
```
说明）。仅获取用户自己的论文和开放获取来源。
切勿仅为规避检测而重写标记段落，却不引用复制的观点——正确的修复方式是添加恰当的引用/引号，或是真正的改写，并将问题告知作者。
与作者自身工作的重叠内容需作为伦理/文本复用问题上报，不得忽略。

Memory

记忆机制

Uses the shared

.paper-memory/

convention (full spec:

paper-memory-convention.md

At start: read
```
lessons.md
```
for prior originality flags so repeat offenders (a section the author keeps recycling) are checked first.
At end: append durable findings —
```
date · check-originality · <scope> · finding · fix
```
— after deduping. A passage recycled across multiple drafts is
```
recurring
```
.
Create
```
.paper-memory/
```
on demand; it is local-only, never uploaded.

遵循共享的

.paper-memory/

约定（完整规范：

paper-memory-convention.md

）。

开始时：读取
```
lessons.md
```
中的过往原创性标记，优先检查作者反复复用的章节。
结束时：去重后添加持久化发现——格式为
```
日期 · check-originality · <范围> · 发现 · 修复方案
```
。若某一段落在多个草稿中被复用，则标记为
```
recurring
```
（重复出现）。
按需创建
```
.paper-memory/
```
；该目录仅存储在本地，绝不会上传。