Datasets API

Loaders for standard Temporal RAG benchmarks.

TEMPO (Primary)

A dataset for temporal retrieval and reasoning across multiple domains (History, Politics, Finance, etc.).

from tempoeval.datasets import load_tempo
ds = load_tempo(domain="history", split="test")

---

TimeQA (Reasoning)

Question Answering dataset requiring explicit temporal constraint satisfaction.

from tempoeval.datasets import load_timeqa
ds = load_timeqa(split="test")

---

SituatedQA

Dataset where answers change depending on the temporal context (e.g., "Who represents District 5?").

from tempoeval.datasets import load_situatedqa
ds = load_situatedqa()

---

TimeBench

Complex temporal reasoning benchmark involving reasoning over timelines.

from tempoeval.datasets import load_timebench

---

Complex TempQA

Multi-hop queries requiring synthesis of multiple temporal facts.

from tempoeval.datasets import load_complex_tempqa

---

Dataset Structure

All loaders return a list of TemporalQuery objects:

@dataclass
class TemporalQuery:
    id: str
    query: str
    gold_ids: List[str]
    year: Optional[int] = None
    domain: Optional[str] = None
    query_guidance: Optional[Dict] = None