fact-checking-methodology - SKILL.md Agent Skill

name: fact-checking-methodology description: Rigorous fact-checking methodology for technology content, ensuring accuracy, credibility, and journalistic standards. Use when verifying claims, statistics, technical assertions, or preparing content for publication.

Fact-Checking Methodology

Обязательные правила вывода

Всегда отвечай на русском.
Сохраняй артефакты в outputs/tech-content-creator/skills/fact-checking-methodology/{timestamp}_{кратко}.md через Write tool; обновляй один файл по итерациям.
Формат: цель/аудитория → бриф → проверки/источники → выводы/решения → TODO → изменения vs прошлой версии.

3-итерационный контур

Диагностика (0.5–1 ч): собрать утверждения/данные/код, источники/версии, риски (легал/бренд/PII). Черновой бриф + риск/decision log.
Проверка (1–2 ч): иерархия источников (Tier1→Tier2→Tier3), технический прогон кода/команд, контекст/даты/версии, поиск контраргументов. Заполни чек-лист.
Верификация (0.5–1 ч): финальный статус (проверено/оспорено/удалено), список источников, дисклеймеры, алерты. Зафиксируй изменения и TODO.

When to Use This Skill

Verifying claims in draft articles before publication
Checking technical accuracy of code examples and configurations
Validating statistics, data, and research citations
Reviewing competitor or third-party claims about technology
Conducting pre-publication review for high-stakes content
Training content creators on verification standards

Fact-Checking Framework

The Three-Layer Verification Model

Layer 1: Source Verification / Проверка источников

Is the source authoritative and credible?
Is the source primary, secondary, or tertiary?
Does the source have potential bias or conflict of interest?
Is the information current and still accurate?

Layer 2: Claim Verification / Проверка утверждений

Can the claim be independently verified?
Do multiple credible sources corroborate?
Is necessary context provided?
Are there counterexamples or exceptions?

Layer 3: Technical Verification / Техническая проверка

Do code examples actually work?
Are version numbers and compatibility accurate?
Are configurations and commands correct?
Are architecture descriptions technically sound?

Source Hierarchy

Tier 1: Primary Authoritative Sources (Highest Credibility)

Official Technical Documentation

Vendor documentation (AWS, Azure, Google Cloud, etc.)
Open source project official docs
API references and specifications
Release notes and changelogs

Peer-Reviewed Research

Academic papers in reputable journals (IEEE, ACM)
Conference proceedings (USENIX, SOSP, OSDI)
arXiv preprints (with caution—not peer-reviewed)

Original Research and Data

Primary research reports (Gartner, Forrester, IDC)
Government and regulatory data
Company financial reports and earnings calls
Original survey data with methodology disclosed

Direct Sources

Official company announcements and press releases
Direct interviews with subject matter experts
First-hand case studies and customer testimonials

Входы (собери до старта)

Перечень утверждений/данных/кодовых примеров, версии/окружения.
Источники (Tier1/2/3), ограничения легал/бренд/PII, дедлайн.

Выходы (обязательно зафиксировать)

Статус по каждому утверждению (подтверждено/оспорено/удалено/дисклеймер).
Список источников/ссылок, результаты прогонов кода, даты проверки.
Дисклеймеры/легал примечания, TODO с владельцами/датами, изменения vs прошлой версии.

Метрики и алерты

Количество исправлений/ретрактов после публикации.
Доля проверенных утверждений и кода, SLA на проверку.
Алерты: сомнительные источники, устаревшие версии, PII/легал риск, несостыковки данных.

Качество ответа (checklist)

Источники ранжированы (Tier1→3), факты подтверждены ≥2 авторитетными, код прогнан.
Даты/версии/контекст указаны, контраргументы рассмотрены.
Дисклеймеры/ограничения присутствуют, логи обновлены, TODO/владельцы/сроки есть.

Red Flags

Опора на Tier3 без Tier1/2, отсутствие контекста/дат/версий.
Непроверенный код/команды, нет подтверждений или ссылок.
Легал/бренд/PII не проверены; нет алертов/чек-листа.

Шаблоны и справочники

Assets: fact-check-worksheet.md, source-log-template.md.
References: source-reliability-tiers.md, fact-check-checklist.md.

Tier 2: Secondary Authoritative Sources (Good Credibility)

Reputable News Organizations

Reuters, Bloomberg, Wall Street Journal, Financial Times
Technology-focused: The Information, Protocol
Russian: РБК, Коммерсантъ, Ведомости

Established Technology Publications

IEEE Spectrum, Communications of ACM
InfoQ, The New Stack
Well-edited tech media with fact-checking: Ars Technica, Wired

Industry Analyst Commentary

Analyst interpretations and predictions (Gartner, Forrester)
Market research firms
Technology advisory firms

Expert Perspectives

Recognized industry experts with track record
University faculty in relevant specialization
Book authors with credentials

Tier 3: Use with Caution (Verify Independently)

Vendor Marketing Materials

Whitepapers (check for bias)
Case studies (verify if possible)
Product comparison documents
Blog posts from vendors

General Technology Media

News aggregators and general tech blogs
Unedited contributed content
Press release republishing

Community Sources

Stack Overflow (good for how-to, not for claims)
Reddit, HackerNews (useful for discussion, not facts)
Personal blogs (unless from recognized expert)

Tier 4: Red Flags (Require Extensive Verification)

Questionable Sources

Anonymous sources without corroboration
Sources with clear conflicts of interest
Clickbait or sensationalist publications
Circular citations (A cites B, B cites A)
Outdated information presented as current
Unsourced statistics ("studies show...")

Verification Checklists

Statistics and Data Checklist

When encountering any statistic, verify:

Source identified: Where does this number come from?
Original source accessed: Not cited secondhand
Methodology disclosed: How was data collected?
Sample size adequate: Large enough to be meaningful?
Date confirmed: When was data collected/published?
Context provided: What do these numbers actually mean?
Definitions clear: What exactly is being measured?
Limitations acknowledged: What doesn't this data tell us?

Example: Verifying "67% of AI projects fail"

✅ Proper Verification:

Find original source: [Gartner research, July 2022]
Check methodology: [Survey of 2,500 enterprises]
Verify definition of "failure": [Didn't reach production within 2 years]
Add context: "According to Gartner's 2022 survey of 2,500 enterprises, 67% of AI initiatives failed to reach production within two years of starting development."

❌ Insufficient Verification: "67% of AI projects fail [source: TechBlog.com article]"

Secondary source, no methodology, no date, no context

Technical Claims Checklist

For any technical assertion:

Test it: If possible, reproduce the claim
Check versions: Version-specific features accurately stated?
Verify compatibility: Integrations work as described?
Validate best practices: Is this actually recommended?
Check deprecations: Is feature still supported?
Confirm limits: Are performance/scale claims accurate?
Review security: Are security implications correct?

Example: Verifying "Kubernetes automatically scales pods"

✅ Accurate with Context: "Kubernetes can automatically scale pods using Horizontal Pod Autoscaler (HPA), which adjusts replica count based on CPU utilization, memory, or custom metrics. However, this requires configuration—autoscaling is not enabled by default."

❌ Misleading: "Kubernetes automatically scales pods" (implies it's automatic without setup)

Code Examples Checklist

For any code snippet in content:

Actually run it: Execute code to verify it works
Complete: Includes necessary imports and context
Correct syntax: No syntax errors
Version-specific: Works with versions mentioned
Best practices: Follows language/framework conventions
Error handling: Handles errors appropriately
Security: No vulnerabilities or insecure practices
Tested: Ideally with automated tests

Example: Python Code Verification

# ❌ UNVERIFIED - May not work
import requests
data = requests.get("https://api.example.com").json()

# ✅ VERIFIED - Complete and working
import requests
from typing import Dict, Optional

def fetch_api_data() -> Optional[Dict]:
    """Fetch data from API with proper error handling."""
    try:
        response = requests.get(
            "https://api.example.com/data",
            timeout=10
        )
        response.raise_for_status()
        return response.json()
    except requests.RequestException as e:
        print(f"API request failed: {e}")
        return None

Quote and Attribution Checklist

For any quote or attribution:

Exact wording: Quote is word-for-word accurate
Proper attribution: Correct name, title, company
Context preserved: Quote meaning not distorted
Recent and relevant: Quote is from appropriate timeframe
Paraphrasing accurate: If paraphrased, meaning preserved
Credentials verified: Person's expertise confirmed
Permission obtained: If needed for case studies

Company and Product Claims Checklist

For claims about companies, products, or market:

Company name correct: Proper spelling and capitalization
Product name current: Not outdated or rebranded
Leadership current: Titles and positions up-to-date
Financials accurate: Revenue, valuation, funding verified
Market position verified: "Leading," "first," "largest" claims checked
Product capabilities accurate: Features exist as described

Common Fact-Checking Scenarios

Scenario 1: Vague Statistics

Claim: "Most companies are adopting Kubernetes"

Issues:

"Most" is undefined (>50%? >75%?)
"Companies" is vague (all companies? Fortune 500? Tech companies?)
No timeframe
No source

Fact-Checked Version: "According to the CNCF's 2023 annual survey, 66% of respondents reported using Kubernetes in production, up from 58% in 2022. The survey included 4,800 respondents primarily from technology and cloud-native companies."

Scenario 2: Outdated Information

Claim: "Docker is the only way to run containers"

Issues:

Was never fully accurate
Increasingly outdated (containerd, CRI-O, Podman)
Doesn't account for OCI standards

Fact-Checked Version: "While Docker pioneered container tooling and remains widely used, modern container ecosystems support OCI-standard runtimes like containerd (used by Kubernetes), CRI-O, and Podman, giving teams multiple options."

Scenario 3: Misleading Comparisons

Claim: "Our service is 10x faster than the competition"

Issues:

Which competitor?
Under what conditions?
What metric (latency, throughput)?
Benchmark methodology?

Fact-Checked Version: "In our benchmark tests using Apache JMeter with 1,000 concurrent users, our service achieved median response times of 45ms compared to 450ms for Service X and 380ms for Service Y. Full methodology and results: [link]"

Scenario 4: Circular Citations

Claim: "95% of data breaches are caused by human error"

Investigation Reveals:

TechBlog A cites Industry Report B
Industry Report B cites News Article C
News Article C cites TechBlog A (circular!)
Original primary source cannot be found

Fact-Checking Action:

Flag as unverifiable
Search for actual research on topic
Either find credible source or remove claim
Lesson: Always trace to primary source

International and Russian Market Fact-Checking

Verifying Russian Market Data

Trusted Russian Sources / Надежные российские источники:

Official Statistics

Росстат (Federal State Statistics Service)
Банк России (Central Bank)
Минцифры (Ministry of Digital Development)

Market Research

РУССОФТ reports
iKS-Consulting
CNews Analytics
Data Insight

News and Analysis

РБК, Коммерсантъ, Ведомости (business press)
Habr (tech community, verify claims)
VC.ru (startup ecosystem)

Challenges:

Less transparency than Western markets in some areas
Russian vs. international methodology differences
Currency and economic data requires careful dating
Regulatory environment changes require recency checks

Cross-Border Verification

When writing for both Russian and international audiences:

Verify statistics apply to claimed geography
Check if international trends mirror Russian market
Note regional differences when relevant
Use locale-appropriate sources for each market

Fact-Checking Tools and Techniques

Search Techniques

Finding Original Sources:

Put statistic in quotes: "67% of AI projects fail"
Search with "original report" or "methodology"
Use site-specific search: site:gartner.com kubernetes adoption
Try date range limits: after:2023-01-01
Reverse image search for charts/infographics

Verifying Claims:

Search for contradictory information
Look for expert commentary or analysis
Check fact-checking sites (Snopes, FactCheck.org)
Verify through multiple independent sources

Documentation Verification

Official Docs:

Always check official documentation for technical claims
Verify version-specific features
Check deprecation notices
Review release notes for changes

GitHub and Open Source:

Check repository for features claimed
Review issues for known problems
Verify version tags and release notes
Check community discussions

Expert Consultation

When claims are hard to verify:

Consult internal subject matter experts
Reach out to technology experts in network
Ask in professional communities (with discretion)
Interview original researchers or practitioners

Handling Unverifiable Claims

When Verification Fails

Option 1: Remove the Claim If claim is central but unverifiable, consider removing or replacing.

Option 2: Add Qualifier "According to [source], ..." or "Based on limited available data..."

Option 3: Acknowledge Limitations "While comprehensive data isn't available, industry observations suggest..."

Option 4: Request Original Source Ask content creator or SME for original source and verification.

Flagging Uncertainty

Be transparent about confidence levels:

✅ High Confidence: "According to AWS documentation..." ✅ Medium Confidence: "Industry analysts estimate..." ✅ Low Confidence: "While exact figures aren't public, estimates range from..." ❌ Unverifiable: [Remove or add strong qualifier]

Fact-Checking Report Template

See references/fact-check-report-template.md for comprehensive bilingual template.

Quick Format:

# Fact-Check: [Article Title]

## Summary
- Claims checked: [X]
- Verified: [Y] ✅
- Need correction: [Z] ⚠️
- Unverifiable: [W] ❌

## Critical Issues
1. [Claim] - [Issue] - [Suggested fix]

## Verification Notes
[Detailed findings and sources]

Best Practices

Fact-Checking Workflow

First read: Mark all factual claims requiring verification
Prioritize: Start with statistics, technical claims, quotes
Trace sources: Always go to primary source
Document: Record verification process and sources
Flag issues: Clearly mark problems with severity
Suggest fixes: Provide corrected versions with sources
Follow up: Verify corrections implemented

Quality Standards

Multi-source rule: Key claims require 2-3 independent sources
Recency rule: Data should be from last 12-24 months (tech moves fast)
Primary source rule: Always verify against original, not secondary citation
Context rule: Provide enough context for readers to interpret claims correctly
Transparency rule: Cite sources so readers can verify themselves

Ethical Considerations

Intellectual honesty: Present data fairly, acknowledge limitations
Avoid cherry-picking: Don't ignore contradictory evidence
Transparent sourcing: Always cite where information comes from
Respect copyright: Don't plagiarize, always attribute
Protect sources: Handle confidential information appropriately

References

references/fact-check-report-template.md - Detailed fact-checking report structure
references/source-evaluation-guide.md - Guide to evaluating source credibility
assets/fact-checking-examples.md - Real examples with explanations

Related Skills

tech-trends-research - Finding reliable sources during research
technical-writing-standards - Writing accurately verified claims