QUICK START

VerityNgn Quick Start Guide

Get up and running with VerityNgn in 5 minutes.

---

Prerequisites

Before you begin, ensure you have:

• ✅ Python 3.12+ installed
• ✅ Google Cloud account
• ✅ Completed Setup Guide (authentication configured)

---

Quick Start

1. Verify Installation

cd /path/to/verityngn-oss
python test_credentials.py

Expected output:

✅ Google Cloud Project: your-project-id
✅ Vertex AI Authentication: SUCCESS
✅ All required credentials configured!

2. Run Your First Verification

Using Streamlit UI (easiest):

./run_streamlit.sh

Then:

1. Open http://localhost:8501 in your browser
2. Paste a YouTube URL
3. Click “Analyze Video”
4. Wait 8-12 minutes for a typical 30-minute video
5. View the interactive report!

Using Command Line:

python -m verityngn.workflows.main_workflow --url "https://www.youtube.com/watch?v=VIDEO_ID"

Using Test Script:

# Test with a specific video
./run_test_tl.sh

3. View Results

Reports are saved to:

outputs/
└── VIDEO_ID/
    ├── report.html      # Interactive HTML report (open in browser)
    ├── report.md        # Markdown summary
    ├── report.json      # Machine-readable structured data
    └── VIDEO_ID_claim_N_sources.html  # Evidence for each claim

Open the HTML report:

# macOS
open outputs/VIDEO_ID/report.html

# Linux
xdg-open outputs/VIDEO_ID/report.html

---

What to Expect

Processing Time

Video Length	Expected Time	Segments	API Calls
10 minutes	8-12 minutes	1	1
20 minutes	8-12 minutes	1	1
33 minutes	8-12 minutes	1	1
60 minutes	16-24 minutes	2	2

Why so fast in v2.0? Intelligent segmentation reduces API calls by 86%!

Progress Logging

You’ll see progress updates like:

🎬 [VERTEX] Segmentation plan: 1998s video → 1 segment(s) of 2860s each
   Expected time: ~8-12 minutes total

🎬 [VERTEX] Segment 1/1: Processing 0s → 1998s (33.3 minutes)
   ⏱️  Expected processing time: 8-12 minutes for this segment
   📊 Progress: 0% complete
   ⏳ Please wait... (this is NORMAL, not hung)

✅ Analysis complete! Processing claims...

Don’t worry if you see no output for 10+ minutes - multimodal analysis takes time!

---

Understanding the Report

Truthfulness Score

Reports include a probabilistic truthfulness distribution:

TRUE:      45%  (Supporting evidence found)
FALSE:     30%  (Contradictory evidence found)
UNCERTAIN: 25%  (Insufficient evidence)

How to interpret:

• High TRUE%: Claims are well-supported by credible sources
• High FALSE%: Claims are contradicted by evidence
• High UNCERTAIN%: Not enough evidence to make determination

Claims Analysis

Each claim shows:

• Claim text: What was claimed in the video
• Specificity score: 0-100 (higher = more verifiable)
• Evidence: Sources supporting or refuting the claim
• Validation power: Source credibility (-1.0 to +1.5)
  • Peer-reviewed: +1.5
  • Scientific source: +1.0
  • Press release: -1.0

Counter-Intelligence

Reports include:

• YouTube Reviews: Debunking or contradictory videos found
• Press Release Detection: Identifies promotional content
• Absence Claims: What’s NOT mentioned (e.g., “No peer-reviewed studies cited”)

---

Common Use Cases

Health & Supplement Claims

# Analyze a health supplement video
python -m verityngn.workflows.main_workflow --url "https://www.youtube.com/watch?v=HEALTH_VIDEO_ID"

VerityNgn excels at:

• Detecting unsubstantiated health claims
• Finding scientific evidence (PubMed, studies)
• Identifying press releases and promotional content
• Discovering contradictory reviews

Product Reviews

# Analyze a product review
python -m verityngn.workflows.main_workflow --url "https://www.youtube.com/watch?v=REVIEW_VIDEO_ID"

VerityNgn checks for:

• Counter-reviews and debunking videos
• Promotional bias detection
• Verifiable claims vs marketing speak

Educational Content

# Verify educational claims
python -m verityngn.workflows.main_workflow --url "https://www.youtube.com/watch?v=EDU_VIDEO_ID"

VerityNgn validates:

• Scientific accuracy
• Source citations
• Expert credentials

---

Tips for Best Results

1. Choose Verifiable Content

Good:

• Product reviews with specific claims
• Health/supplement marketing videos
• Educational content with factual claims

Less Suitable:

• Opinion/commentary videos
• Entertainment/fiction content
• Artistic performances

2. Allow Sufficient Time

• Don’t interrupt during “Processing segment…” messages
• 10+ minutes of no output is NORMAL for large segments
• Check progress logs for status updates

3. Interpret Results Carefully

• High UNCERTAIN% doesn’t mean FALSE - it means “not enough evidence”
• Check validation power of sources (peer-reviewed > general web)
• Review counter-intelligence findings for context

4. Use Full API Setup for Best Results

Without Google Custom Search API:

• ✅ Video analysis works
• ✅ Claims extraction works
• ⚠️ Limited evidence verification

With Google Custom Search API:

• ✅ Full evidence gathering
• ✅ Credible source validation
• ✅ Scientific database integration

See Setup Guide - API Keys

---

Troubleshooting

”Process seems hung”

Normal behavior: 8-12 minutes of no output during segment processing

How to check if it’s really hung:

1. Look for last log message: “Processing segment…”
2. Check timestamp - has it been > 15 minutes?
3. If yes, check network connection and API quotas

”Empty response from analysis”

Possible causes:

1. Network timeout - retry with smaller video
2. API quota exceeded - check Google Cloud quotas
3. Model error - check logs for specific error

Solution: Try a shorter video first (< 10 minutes)

“Permission denied”

Cause: Service account lacks Vertex AI permissions

Solution:

gcloud projects add-iam-policy-binding YOUR-PROJECT-ID \
    --member="serviceAccount:SERVICE-ACCOUNT-EMAIL" \
    --role="roles/aiplatform.user"

See Setup Guide - Troubleshooting

---

Next Steps

Learn More

• Testing Guide - Run comprehensive tests
• Architecture - Understand how it works
• Configuration - Advanced settings
• Research Papers - Read the methodology

Try Advanced Features

• Custom segmentation: Override automatic segment calculation
• Batch processing: Analyze multiple videos
• API integration: Use VerityNgn programmatically

Contribute

• Report issues on GitHub
• Suggest improvements
• Share your use cases

---

Example Workflow

Here’s a complete example analyzing a 33-minute health supplement video:

# 1. Navigate to project
cd /Users/ajjc/proj/verityngn-oss

# 2. Activate environment
conda activate verityngn

# 3. Run analysis
python -m verityngn.workflows.main_workflow \
    --url "https://www.youtube.com/watch?v=tLJC8hkK-ao"

# 4. Wait ~10 minutes...

# 5. View report
open outputs/tLJC8hkK-ao/report.html

Expected output:

• Claims extracted: 15-25
• Processing time: 8-12 minutes
• Evidence sources: 50-100
• Counter-intel findings: 2-5 YouTube reviews

---

Need help? See Setup Guide or Troubleshooting

Last Updated: October 28, 2025
Version: 2.0