paper-signal-audit

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Paper Signal Audit

Paper Signal 审核

Judge the image that exists, not the prompt that intended it.

以实际存在的图像为判断依据，而非生成图像的提示词。

Load the audit basis

加载审核基准

Read from

../paper-signal/

```
references/subject-routing.md
```
```
references/visual-language.md
```
```
references/quality-contract.md
```
```
references/restyle-contract.md
```
for edits

Read the project prompt, manifest, preservation record, outline, and evidence ledger when available.

从

../paper-signal/

目录读取以下文件：

```
references/subject-routing.md
```
```
references/visual-language.md
```
```
references/quality-contract.md
```
若为编辑类任务，读取
```
references/restyle-contract.md
```

若有可用文件，同时读取项目提示词、清单、保留记录、大纲及证据台账。

Audit workflow

审核流程

Inspect every local image with
```
view_image
```
at original detail.
Identify the delivery route, subject route, and intended composition mode from artifacts; infer them explicitly when artifacts are missing.
Compare the raster against protected subject details, exact text, factual claims, and requested aspect ratio.
Inspect at full size for identity, geometry, paper, image integration, type, print defects, and AI artifacts.
Inspect at thumbnail size for subject recognition, chromatic signal, hierarchy, and series recognition.
For a series, compare the whole sequence for material continuity, composition variety, pacing, and cloned-layout drift.
Rank findings by severity and give one correction path per failed image.

使用
```
view_image
```
工具以原始细节精度检查每张本地图像。
从图像痕迹中识别交付路径、主题路径及预期构图模式；若没有明显痕迹，则明确推断这些信息。
将光栅图像与受保护的主题细节、精确文本、事实声明及要求的宽高比进行对比。
以全尺寸检查图像的辨识度、几何形态、纸张质感、图像融合度、字体、印刷缺陷及AI生成痕迹。
以缩略图尺寸检查主题可识别性、色彩信号、层级结构及系列辨识度。
对于系列作品，对比整个序列的材质连贯性、构图多样性、节奏及克隆布局偏差。
按严重程度对检测结果排序，并为每张不合格图像提供一条修正路径。

Report format

报告格式

Lead with the verdict:

```
Pass
```
— ready to use
```
Pass with refinements
```
— usable; small craft changes would improve it
```
Regenerate
```
— one or more blocking failures

Then report only observed issues:

Severity	Image	Observation	Why it matters	Targeted correction

Separate:

blocking: subject drift, false/wrong text, wrong ratio, unreadable essential content, missing output
major: generic template composition, weak subject scale, incoherent print process, pseudo-language, muddy hierarchy
minor: slightly weak signal, texture repetition, edge or safe-zone refinement

Do not praise or criticise with unsupported adjectives. Point to visible evidence.

开头先给出结论：

```
Pass
```
— 可直接使用
```
Pass with refinements
```
— 可使用；小幅工艺调整可提升质量
```
Regenerate
```
— 存在一个或多个阻塞性缺陷

随后仅报告观察到的问题：

严重程度	图像	观察结果	影响原因	针对性修正方案

按以下类别区分问题：

阻塞性：主题偏差、文本错误/虚假、宽高比错误、核心内容不可读、输出缺失
主要：通用模板构图、主题比例失衡、印刷工艺不一致、伪语言、层级结构模糊
次要：信号稍弱、纹理重复、边缘或安全区域需优化

不得使用无依据的形容词进行褒贬。需指向可见证据。

Anti-AI diagnosis

反AI生成诊断

Check specifically for:

meaningless microtext or faux Asian characters
generic birds, moons, stairs, doors, flowers, and registration marks unrelated to content
uniformly applied vintage noise
beige template plus clean modern typography
perfect vector spacing under fake paper texture
waxy faces, altered identity, impossible hands, warped buildings, or invented product details
random collage fragments with no visual thesis
empty space that creates no direction or tension

重点检查以下特征：

无意义的微文本或仿亚洲字符
与内容无关的通用元素，如鸟类、月亮、楼梯、门、花朵及定位标记
均匀添加的复古噪点
米色模板搭配简洁现代字体
虚假纸张纹理下的完美矢量间距
蜡质感面部、身份篡改、畸形手部、扭曲建筑或虚构产品细节
无视觉主题的随机拼贴碎片
无法营造方向感或张力的空白区域

Correction policy

修正规则

Correct one failure class per regeneration. Preserve accepted subject features, text, material choices, and crop. Save the correction prompt and output as versioned siblings.

When the user asks only for review or diagnosis, stop after the report and correction prompt. Generate or edit the image only when the user explicitly asks to fix, remake, or produce it.

Never patch generated text or composition with code.

每次重新生成仅修正一类缺陷。保留已认可的主题特征、文本、材质选择及裁剪方式。将修正提示词及输出结果保存为同版本的关联文件。

当用户仅要求审核或诊断时，完成报告及修正提示后即停止操作。仅当用户明确要求修复、重制或生成图像时，才进行图像生成或编辑操作。

禁止使用代码修补生成的文本或构图。