15,000 JSONL examples of fact-based question-and-answer pairs created by WithinUsAI. Each answer includes cited fact sources and a verification tag. The dataset was last updated on January 4, 2026.
Use Cases
- Train tiny models for factual question answering based on cited sources
- Benchmark model performance on verifiable truthfulness based on verification tags
- Develop constraint-aware language models optimized for short contexts based on TRM-native prompts
Strengths
- 15,000 examples provide a substantial training corpus
- Each answer includes fact_sources (citations) for verifiable information
- Prompts are described as optimized for tiny contexts
Limitations
- Column-level documentation is absent; field semantics must be inferred after download
- Row count is known, but specific data content and quality require manual inspection
Provenance
- Source
- WithinUsAI
- Freshness
- Last updated 2026-01-04 00:09:22