---
format: "aidr-story-markdown/v1"
id: "b7d75266f87b25ddf9315a1a37bb3d424673d794fb0adb53a2468b29593ad0f7"
canonical_url: "https://aidr.today/b7d75266?lang=en"
title: "Trajectory2Gym: Decision-Grounded Compositional Interaction Synthesis for Tool-Use Agents"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-27T00:00:00.000Z"
category: "Research"
topics: ["agent"]
source_urls: ["https://huggingface.co/papers/2610.33568"]
summary: "Trajectory2Gym is a method for generating tool-use agent training data that emphasizes compositional coverage and decision grounding, constructing executable interactions around under-supported structures and deriving supervision only from grounded decisions in their realized executions."
---

# Trajectory2Gym: Decision\-Grounded Compositional Interaction Synthesis for Tool\-Use Agents

> [Open the canonical story](<https://aidr.today/b7d75266?lang=en>)

**Published:** 2026-10-27T00:00:00.000Z
**Category:** Research
**Topics:** agent

## Summary

Trajectory2Gym is a method for generating tool\-use agent training data that emphasizes compositional coverage and decision grounding, constructing executable interactions around under\-supported structures and deriving supervision only from grounded decisions in their realized executions\.

## Sources

- [Story source](<https://huggingface.co/papers/2610.33568>)