Ambiguity Resolution Corpus for SQL. 311 task instances across 6 databases.
tasks.jsonl — one row per intended interpretation.
databases/sqlite/*.sqlite — databases referenced by the db field.
databases/column_meanings.json — column-level descriptions.
from datasets import load_dataset
ds = load_dataset("kiaw73b5o7a/arcs-bench", split="train")
Most rows have the gold SQL string (gold_queries[*].query) masked to null to avoid label leakage;… See the full description on the dataset page:
https://huggingface.co/datasets/kiaw73b5o7a/arcs-bench.
