Atomic-Claim CNN/DailyMail: Synthetic Atomic Claims for Fact-Checking

A high-quality synthetic dataset of 1,096 atomic factual claims extracted from CNN/DailyMail news articles using Gemini 2.5 Pro as a synthetic teacher. Designed for training claim extraction models in automated fact-checking pipelines. Key Features: 1,096 article-claims pairs annotated with strict extraction rules Context decoupling: pronouns resolved to full entities Verifiability filtering: only factual statements with specific entities/numbers Attribution enforcement: quotes and statements properly sourced 90/10 train/test split (996 training, 100 test samples) JSONL format for easy streaming Generation Method:All annotations generated using a fixed, version-controlled Gemini 2.5 Pro prompt template without manual post-editing, ensuring full reproducibility. Use Cases:- Training claim extraction models for fact-checking systems- Teacher-student knowledge distillation for small language models- Benchmarking claim detection pipelines- RAG-based verification system development