---
type: "post"
id: "turning-database-grading-into-rl-rewards-post-training-46f950"
slug: "turning-database-grading-into-rl-rewards-post-training-46f950"
title: "Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox"
description: "Microsoft and Hugging Face published ThinkingBox-Training, a post-training setup that uses executable checks of a database’s final state as reinforcement-learning rewards for agents."
retrieval_nugget: "Microsoft and Hugging Face published ThinkingBox-Training, a post-training setup that uses executable checks of a database’s final state as reinforcement-learning rewards for agents."
published_at: "2026-10-22"
updated_at: "2026-10-22"
record_date: "2026-10-22"
date_kind: "published_at"
topics: ["coding-agents","evals"]
entities: ["Microsoft","Hugging Face","ThinkingBox","ThinkingBox-Bench","Qwen3.8-27B","GRPO"]
related_patterns: ["verification-bandwidth-is-the-scarce-resource","goal-scoped-loops-replace-manual-continuation"]
novelty_label: "new_fact"
what_changed: "Microsoft and Hugging Face report ThinkingBox database-state checks as RL rewards, lifting Qwen3.8-27B success from 51.70% to 60.78% after 50 GRPO steps."
prior_refs: ["memory:852d4b1b-4088-8546-adfc-7c802fafc1f7"]
source_url: "https://huggingface.co/blog/microsoft/thinkingbox-training"
source_urls: ["https://huggingface.co/blog/microsoft/thinkingbox-training"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:turning-database-grading-into-rl-rewards-post-training-46f950"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/turning-database-grading-into-rl-rewards-post-training-46f950/","markdown":"https://newruntime.com/posts/turning-database-grading-into-rl-rewards-post-training-46f950.md","json":"https://newruntime.com/posts/turning-database-grading-into-rl-rewards-post-training-46f950.json"}
---

# Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox

## Retrieval answer

Microsoft and Hugging Face published ThinkingBox-Training, a post-training setup that uses executable checks of a database’s final state as reinforcement-learning rewards for agents.

Microsoft and Hugging Face published ThinkingBox-Training, a post-training setup that uses executable checks of a database’s final state as reinforcement-learning rewards for agents. The accompanying implementation is available on GitHub, according to the joint Hugging Face post.

ThinkingBox-Bench contains 507 tasks. Each task starts a backend, gives an agent MCP tools and a simulated user with private context, then checks database records after the run. The reward is based on whether the required terminal state is correct, rather than whether the agent followed a prescribed tool-call sequence.

The authors selected training tasks outside the benchmark using a one-time estimate of difficulty for the base policy, then fine-tuned Qwen3.8-27B with GRPO for 50 steps. On the 507-task held-out benchmark, pass@1 rose from 51.70% to 60.78%. Tasks that passed all 20 recorded attempts rose from 38 to 107, while tasks solved at least once fell from 453 to 446.

The source reports a trade-off toward repeatable success rather than broader at-least-once coverage. It does not establish independent replication, general performance beyond this benchmark, or total deployment economics: its cost comparison excludes training and policy-probe costs.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Coding Agents](https://newruntime.com/topics/coding-agents.md): Explore the coding-agents topic hub.
- topic hub: [Evals](https://newruntime.com/topics/evals.md): Explore the evals topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)