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%
01

Overview

drop agent transcript screenshot · 1200×500

The task

The agent receives a natural-language analytics question, such as "Which three product categories had the largest drop in revenue between Q1 and Q2 2024?", and a read-only SQLite database it has never seen. It must explore the schema with tools, write queries, and submit the exact result rows.

Why this environment

Rewards are verifiable: the grader compares result sets, so there is no LLM judge to game. The multi-turn budget rewards schema exploration and self-correction, not one-shot guessing.

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