Reward Atlas
All field guidesEsc to close
Sign in
LLM / AgenticPublicPublishedv1.0Advanced

SQL-Detective-v1

A multi-turn agent answers analytics questions by querying a read-only SQLite sandbox. Tool use, verifiable rewards, 12-turn budget. Maintained by Bhaskar & Jaswanth · updated 28 Aug 2026.

Dataset
2,400 q · 12 DBs
Action
tool call | final answer
Turn cap
12
Best baseline
GRPO-7B 61.4%
04

Action space

Tools

Each turn the policy emits exactly one tool call. Malformed calls count as invalid queries.

toolsignaturereturnsguardrail
list_tables()Table names
describe_table(table: str)Columns and typesUnknown table → error
run_query(sql: str)Up to 200 rowsRead-only, 5 s timeout
submit_answer(rows: list[list])Ends episodeGraded once
Guide details
Version
Type
LLM / Agentic
API
OpenEnv
License
Apache-2.0
Seeds
3
Domains
tool-usedatacoding
Install
pip install reward-atlas-sql-detective

Issue with this step? Suggest an edit