That toolkit above is your direct line to the TREC 2025 DRAGUN track resources. It's the first standardized framework for evaluating AI systems that help readers assess news credibility.

Researchers just released 1,200+ news topics with source documents and human-written reference reports. This isn't just academic—it's the missing benchmark for building RAG systems that actually fight misinformation.

Why This Toolkit Matters Now

Every day, readers face a firehose of conflicting news. Traditional fact-checking can't scale. AI assistants could help, but until now, there was no way to properly evaluate them.

The DRAGUN (Detection, Retrieval, and Augmented Generation for Understanding News) track fixes this. It provides:

• 1,200+ complex news topics from real-world events
• Source documents with varying reliability levels
• Human-written reference reports showing ideal responses
• Automated scoring metrics for attribution and completeness

How The Evaluation Works

Your AI system takes a news topic and related documents. It generates a reader-friendly report explaining trustworthiness. The toolkit then scores it on three critical dimensions:

Attribution Accuracy: Does the report correctly cite its sources? This prevents hallucination—the AI making up facts.

Content Completeness: Does it cover all key aspects readers need? Missing crucial context is as bad as being wrong.

Helpfulness: Is the report actually useful for decision-making? Technical accuracy means nothing if readers can't understand it.

The Real-World Impact

This isn't just for researchers. Developers can now build better browser extensions, news aggregators, and educational tools.

Imagine an AI sidebar that automatically generates credibility reports for any news article. Or a learning tool that shows students how to evaluate sources. The DRAGUN resources make these possible with proper evaluation.

The scoring system uses both automated metrics and human judgments. This combination catches nuances that pure automation misses.

Getting Started With Your Own System

Use the commands in the Quick-Value Box to download everything. The dataset includes JSON files with topics, documents, and reference reports.

Build your RAG system to generate reports. Then run the scoring scripts to see how it performs. The metrics will show exactly where your system needs improvement.

This is iterative development with clear feedback. You're not building in the dark anymore.