Skip to content
Back to skills

2566 Quality Rubric 5c8b32c0

ASecurity

| Criteria | Examples | |----------|----------| | Peer-reviewed systematic reviews | Cochrane Reviews, PRISMA-compliant studies | | Meta-analyses | Statistical aggregations of multiple studies | | Randomized controlled trials | Clinical trials, A/B testing at scale | | Official government publications | FDA, NIH, WHO, regulatory bodies | | Major institution research | MIT, Stanford, Google Research, OpenAI |

  • 9 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 11, 2026
researchrustgotestinggitapiperformancedocumentation

Works with

  • cli
  • api

Security analysis

A100/100

Scanned October 11, 2026

npx -y skills add tools-only/X-Skills --skill 2566-quality_rubric_5c8b32c0 --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of 2566 Quality Rubric 5c8b32c0?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for 2566 Quality Rubric 5c8b32c0
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/tools-only-2566-quality-rubric-5c8b32c0/badge)](https://www.skillsdirectory.com/skills/tools-only-2566-quality-rubric-5c8b32c0)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

SKILL.md
# Source Quality Rubric

## Rating System (A-E)

### A - Highest Quality (Primary Sources)
| Criteria | Examples |
|----------|----------|
| Peer-reviewed systematic reviews | Cochrane Reviews, PRISMA-compliant studies |
| Meta-analyses | Statistical aggregations of multiple studies |
| Randomized controlled trials | Clinical trials, A/B testing at scale |
| Official government publications | FDA, NIH, WHO, regulatory bodies |
| Major institution research | MIT, Stanford, Google Research, OpenAI |

**Trust Level**: Can be cited as definitive evidence  
**Verification**: Check DOI, publication venue, author credentials

---

### B - High Quality (Authoritative Sources)
| Criteria | Examples |
|----------|----------|
| Peer-reviewed original research | Journal articles with methodology |
| Official standards documents | ISO, IEEE, W3C specifications |
| Established organization white papers | Gartner, McKinsey, Big Tech research |
| Clinical/practice guidelines | Medical associations, professional bodies |
| Official documentation | Product docs, API references |

**Trust Level**: Strong evidence, verify with A-level if critical  
**Verification**: Check author affiliation, funding disclosure

---

### C - Moderate Quality (Expert Sources)
| Criteria | Examples |
|----------|----------|
| Expert opinion pieces | Named experts with credentials |
| Conference presentations | Academic/industry conferences |
| Case studies | Documented implementations |
| Reputable news analysis | NYT, WSJ, BBC technology sections |
| Industry analyst reports | Paid reports with methodology |

**Trust Level**: Useful for context, verify key claims  
**Verification**: Check author expertise, potential bias

---

### D - Lower Quality (Supporting Sources)
| Criteria | Examples |
|----------|----------|
| Preprints | arXiv, bioRxiv (not peer-reviewed) |
| Expert blog posts | Personal blogs from domain experts |
| Conference abstracts | Preliminary findings |
| Press releases | Company announcements |
| Trade publications | Industry news sites |

**Trust Level**: Use as leads, verify before citing  
**Verification**: Cross-reference with higher-quality sources

---

### E - Use with Caution (Weak Sources)
| Criteria | Examples |
|----------|----------|
| Anonymous sources | Unverifiable claims |
| Opinion without evidence | Hot takes, speculation |
| Outdated information | >3 years for fast-moving fields |
| Clear bias/conflicts | Vendor marketing, advocacy sites |
| Social media posts | Twitter/X threads, Reddit comments |
| AI-generated content | Unverified LLM outputs |

**Trust Level**: Do not cite as evidence  
**Verification**: Only use if no alternatives exist, flag uncertainty

---

## Quality Assessment Checklist

For each source, evaluate:

### Authorship
- [ ] Author clearly identified
- [ ] Author has relevant credentials
- [ ] Author affiliation disclosed
- [ ] No undisclosed conflicts of interest

### Publication
- [ ] Publication date clearly stated
- [ ] Publication venue reputable
- [ ] Peer review or editorial process
- [ ] Still current/relevant

### Content
- [ ] Methodology described (if research)
- [ ] Data sources cited
- [ ] Claims supported by evidence
- [ ] Limitations acknowledged

### Accessibility
- [ ] URL accessible and stable
- [ ] DOI available (for academic)
- [ ] Archive link created (for ephemeral content)

---

## Domain-Specific Adjustments

### Technology/AI
- Preprints may be acceptable (field moves fast)
- Company research blogs (Google AI, OpenAI) often quality
- GitHub repositories count as primary sources for code
- Benchmark papers essential for performance claims

### Medical/Health
- Peer-review mandatory for health claims
- Clinical trial registration required
- FDA/regulatory status must be verified
- Patient population and sample size matter

### Business/Market
- Market size claims need methodology
- Analyst reports vary in quality
- Company financials from SEC filings (A-level)
- Press coverage alone insufficient

### Legal/Policy
- Primary legislation and regulations (A-level)
- Court documents and rulings (A-level)
- Legal analysis from law firms (B-level)
- News coverage of legal matters (C-level)

---

## Minimum Quality Thresholds

| Content Type | Minimum Rating |
|--------------|----------------|
| Definitive claims | A or B |
| Statistics/data | A, B, or C with verification |
| Expert opinions | C or higher |
| Background context | D acceptable |
| Future predictions | C with attribution |

---

## Red Flags - Automatic Quality Reduction

- No author attribution → E
- No date → Reduce by 2 levels
- Broken citations → Reduce by 1 level
- Known retraction → E
- Predatory journal → E
- SEO-optimized content farm → E
- Circular citations (sources cite each other) → Investigate

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.