LakeQuest

A Benchmark for Grounded Question Answering over Data Lakes

What is LakeQuest?

LakeQuest evaluates systems that answer natural-language questions grounded in heterogeneous data lakes. The benchmark spans three domains: AIML, bank, and drug.

Systems are tested on answer correctness and, where applicable, evidence grounding over the source lake artifacts.

Getting started

The public starter kit contains the participant-facing submission format, sample predictor, and local validation utilities.

Starter Kit

Benchmark Data

Scoring Program

Benchmark tracks

LakeQuest is hosted as three Codabench tracks because Codabench phases inside one competition must be sequential.

LakeQuest Open Dev

LakeQuest Open Test

LakeQuest Closed Test

Stay connected

@inproceedings{solodko2026lakequest,
  title = {{LakeQuest}: A Three-Domain Benchmark for Grounded Question Answering across Data Lakes},
  author = {Solodko, Michael and Gong, Steven and Yu, Guangwei and Gorti, Satya Krishna and Cresswell, Jesse C. and Zhong, Victor},
  booktitle = {Proceedings of the Conference on Language Modeling (COLM)},
  year = {2026}
}

Benchmark Splits

LakeQuest uses separate open development, open test, and closed held-out code-evaluation tracks.

Lake Open Dev Open Test Closed Test
AIML1,7271,7251,725
Bank1,0211,0201,020
Drug1,1511,1501,150
Total3,8993,8953,895

Leaderboard

Official scores are maintained on Codabench. The table below mirrors the public metric layout.

Open Test
# Model Team Overall Answer F1 Coverage
- Official submissions Codabench View live leaderboard

Evaluation

Prediction submissions contain one predictions.jsonl file at the archive root. Each line must include a question ID and answer text.

{"qa_id": "example-id", "answer": "answer text"}

The main public score emphasizes answer quality. Optional evidence fields may be used for diagnostic grounding metrics when enabled by the organizers.

Tracks

Track Questions Answers Submission
Open Dev Released Released Prediction zip
Open Test Released Hidden Prediction zip
Closed Test Hidden Hidden Runnable project zip