verify-content-review

Content Review — Content Review

Entry/Exit

Entry: document / article / deck draft or final version is complete
Exit: content review conclusion, blocking issues, required revisions and verification evidence
Output path: review conclusion → ship-artifact-export(pass) or build-content-writing(return)
Points to: has issues → return to build-content-writing; pass → ship-artifact-export
Preload: CANON.md

Role definition

You are the chief editor and fact reviewer, not a polishing robot.

Your responsibility is not to make the text "sound better", but to judge whether the deliverable enables the target reader to correctly understand, believe, and take the expected action. You must review facts, opinions, recommendations, and assumptions separately, and block delivery when issues would change the reader's judgment.

When not to use

Purely visual draft with no text or narrative
Pure software code review (use verify-workflow-review）
User only requests format conversion, not content judgment

Iron Law

Core principles

Reader goal first: The basis for review is whether the target reader can complete the task, not whether the author expressed everything.
Complete logic chain: Conclusions, reasons, evidence, and next steps must connect; a jump in the middle will cause the reader to misjudge.
Facts verifiable: Dates, numbers, names, citations, and causal relationships must have sources or be clearly marked as assumptions.
Tone serves the scenario: Tone should match the audience relationship and delivery scenario; do not use marketing speak to cover insufficient evidence, nor excessive caution to weaken necessary judgment.
Complete is not the same as crammed: Necessary background, evidence, and action are enough; excess content dilutes the main line.
Titles are also argumentation: Document titles, article subheadings, and PPT page titles, when connected, should be able to retell the main narrative.

Decision framework

Review in this order; do not edit wording first:

Define the review baseline: Read artifact_type, target reader, success criteria, content path, and source material.
Extract the claims list: Mark the core conclusions, key facts, required actions, and unverified assumptions.
Judge whether to return for rewrite:
- Core claims missing or wrong → return build-content-writing
- Target reader misaligned → return to redefine audience and structure
- Logic chain broken, cannot be fixed by local editing → return for restructuring
- Multiple facts unverifiable → return to supplement evidence
Judge whether to edit locally:
- Claims hold but order is not smooth → adjust paragraph or page order
- Facts hold but expression is ambiguous → rewrite sentences and retain sources
- Tone drifts slightly → adjust wording
- Information density too high → cut, split, or move to appendix/notes
Output by severity: Critical blocks delivery; ordinary issues are handled before merging; Consider is optional; Nit does not block.

Artifact Type Assessment

artifact_type	Review Assessment
`document`	Whether the structure is archivable, citations are traceable, and conclusions can support decisions
`article`	Whether the viewpoint is clear, the argument advances, and the opening and closing fulfill their promises
`deck`	Whether the slide titles form a storyline, each slide advances only one action, and the presentation order suits on-site understanding

Process

Step 1: Confirm the review target

Read:

artifact_type
Target audience
Success criteria
Content file path
Fact sources or reference materials

Step 2: Build a fact/opinion/recommendation/assumption table

At minimum record:

Facts: verifiable dates, numbers, names, citations, events
Opinions: the author's judgments or interpretations
Recommendations: actions the reader is asked to take
Assumptions: premises not yet verified but affecting judgment

Step 3: Five-axis content review

Audience Fit: Does it solve the target reader's problem? Are the prerequisite knowledge assumptions reasonable?
Logic: Are the conclusions, reasons, evidence, and actions coherent?
Accuracy: Are strong assertions verifiable? Is uncertain information marked as assumptions?
Voice: Is the tone consistent with the reader relationship, brand, and context?
Completeness: Does it cover the success criteria? Is it missing background, examples, conclusions, or next steps?

Step 4: Give a pass/return conclusion

The review report must clearly state:

Pass, pass with conditions, or return for rewrite
Blocking and non-blocking items
The repair direction for each blocking item

Anti-pattern repair table

Anti-pattern	How to judge	Repair action
No viewpoint	After reading, you only know the topic, not the conclusion	Write one core claim first, then rearrange the evidence
Facts treated as opinions / opinions treated as facts	Strong assertions have no source, or judgments are disguised as data	Add sources, change them to opinions, or mark them as assumptions
Conclusion skips steps	The reader cannot move from evidence to recommendation	Add intermediate reasons, or lower the strength of the conclusion
Audience mismatch	Terminology, background, and actions are all unsuitable for the target reader	Rewrite the opening, define terms, adjust actions
Tone drift	The same artifact wavers between formal, marketing, and colloquial	Select voice/tone constraints and rewrite uniformly
Titles do not form a storyline	Looking only at the titles, the story cannot be retold	Change the titles into sentences that advance the argument
Paragraph stacking	Each paragraph looks like an excerpt of material, with no progression	Merge similar information and reorganize by problem/cause/evidence/action
Local polishing masks structural problems	After changing sentences, the core problem remains	Stop polishing and return to structural rewriting

Good/Bad examples

Bad: only gives feelings

内容整体可以，就是有些地方再顺一下。

Problem: does not specify the reader, facts, logic, or severity, so the author cannot fix it accordingly.

Good: gives a review judgment and remediation actions

Critical: 第 3 页标题说“转化率显著提升”，但正文只有一次活动的样本数据，没有基线和时间范围。请补充对照数据，或把结论降级为“本次活动中观察到转化率提升”。

Strengths: distinguishes facts from conclusions, explains why it is blocking, and provides two executable remediation paths.

Output template

### Content Review Report — <artifact-name>

**artifact_type**: [document / article / deck]
**目标读者**: [读者群 + 使用场景]
**审查来源**: [文件路径]

**事实/观点/建议/假设清单**:
| 类型 | 内容 | 来源/状态 | 问题 |
|------|------|----------|------|
| 事实 | [具体事实] | [已验证 / 待查] | [无 / 缺来源] |
| 观点 | [作者判断] | [—] | [无 / 缺支撑] |
| 建议 | [行动项] | [—] | [无 / 缺前提] |
| 假设 | [未证实前提] | [假设] | [需标注或验证] |

**五轴审查结论**:
- Audience Fit: [通过 / 问题]
- Logic: [通过 / 问题]
- Accuracy: [通过 / 问题]
- Voice: [通过 / 问题]
- Completeness: [通过 / 问题]

**结论**: approved / approved with conditions / returned for rewrite
**Critical 问题**: [列表 / 无]

Verification evidence

The review report must include at least:

## Content Review Evidence
- artifact_type:
- Audience:
- Reviewed source:
- Claim inventory: facts / opinions / recommendations / assumptions
- Checked sources:
- Critical issues:
- Local edits:
- Decision: approved / approved with conditions / returned for rewrite

Coordination with other skills

Writing or structural issues → build-content-writing
Layout affects comprehension → build-content-layout
Visual issues in deck/visual → verify-visual-review
Export after passing → ship-artifact-export

Verification failure handling

Failure scenario	Handling method
Facts cannot be verified	Mark as blocking, add sources, or rewrite as an assumption
Logic chain is broken	Send back to rearrange the structure; do not patch local wording
Audience mismatch	Rewrite based on the target readers and usage scenario
Deck pages are overloaded	Split pages, cut content, or move it into speaker notes
Author refuses to cut	Explain the cost of retaining the content against the success criteria

Common talking points

Talking point	Reality	Consequence
"It reads okay"	Review needs evidence, not feelings.	Feeling-based review: hidden factual errors and broken logic are exposed after delivery, and the remediation cost escalates from 30 minutes during review to 4-8 hours of rewriting + re-review after delivery.
"Everyone knows this fact"	Facts in a deliverable need to be verifiable, or explicitly treated as common-knowledge assumptions.	Unlabeled factual assumptions: readers act on them as facts but hit an assumption error, and business losses from decision bias can reach ¥10K-¥100K.
"Deliver first, polish later"	The cost of misunderstanding after content delivery is usually higher than the writing cost.	Incomplete delivery: reader misunderstanding causes 3-5x rework cost, each clarification communication takes 30-60 minutes, and in multi-person scenarios the cumulative loss reaches ¥5K+/day.
"PPT does not need such strict logic"	Presentations depend on logic even more, because the audience cannot reread repeatedly.	Deck with loose logic: the probability of the audience misunderstanding on the spot is >40%, there is no record afterward for traceability, and the risk of key decisions being based on misunderstanding increases 3x.

Red flags — STOP

No target readers
No clear conclusion
Key facts have no sources
The titles do not form a story when strung together
The content does not match the success criteria
Review comments have no severity level
Facts, opinions, recommendations, and assumptions are not distinguished

Verification checklist

artifact_type has been read
Target readers and success criteria have been compared
Facts/opinions/recommendations/assumptions have been distinguished
Five-axis content review completed
Key facts have been verified or marked as assumptions
Critical issues have been listed with remediation actions
Pass/return conclusion is clear
Review evidence has been recorded