Metrics¶
TempoEval metrics measure temporal quality across retrieval, generation, and reasoning.
Layers¶
- Retrieval metrics evaluate whether the right time periods are retrieved.
- Generation metrics evaluate whether answers are faithful and aligned in time.
- Reasoning metrics evaluate ordering, durations, and cross-period synthesis.
- Composite metrics combine multiple signals into one score.
Evaluation modes¶
- Gold labels (ID-based relevance)
- Focus Time (QFT/DFT/AFT overlap)
- LLM-as-judge (prompted evaluation)