# TRACE: Training Reasoning Agents for Causal Exploration with Synthesized Rewards

arXiv: https://arxiv.org/abs/2609.10315 · HTML: https://mlsanity.com/paper/2609.10315/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 28 references that are in the map.
- First posted on arXiv 9 Sep 2026.

## Builds on (most cited first, 23 of 23)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [DAPO: An Open-Source LLM Reinforcement Learning System at Scale](https://mlsanity.com/paper/2503.14476/) (2025)
- [Self-Instruct: Aligning Language Models with Self-Generated Instructions](https://mlsanity.com/paper/2212.10560/) (2022)
- [Understanding R1-Zero-Like Training: A Critical Perspective](https://mlsanity.com/paper/2503.20783/) (2025)
- [Spider: A Large-Scale Human-Labeled Dataset for Complex and Cross-Domain Semantic Parsing and Text-to-SQL Task](https://mlsanity.com/paper/1809.08887/) (2018)
- [Scaling Laws for Reward Model Overoptimization](https://mlsanity.com/paper/2210.10760/) (2022)
- [Group Sequence Policy Optimization](https://mlsanity.com/paper/2507.18071/) (2025)
- [MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models](https://mlsanity.com/paper/2309.12284/) (2023)
- [Can LLM Already Serve as A Database Interface? A BIg Bench for Large-Scale Database Grounded Text-to-SQLs](https://mlsanity.com/paper/2305.03111/) (2023)
- [Seq2SQL: Generating Structured Queries from Natural Language using Reinforcement Learning](https://mlsanity.com/paper/1709.00103/) (2017)
- [WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instruct](https://mlsanity.com/paper/2308.09583/) (2023)
- [RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback](https://mlsanity.com/paper/2309.00267/) (2023)
- [MLAgentBench: Evaluating Language Agents on Machine Learning Experimentation](https://mlsanity.com/paper/2310.03302/) (2023)
- [Causal Reasoning and Large Language Models: Opening a New Frontier for Causality](https://mlsanity.com/paper/2305.00050/) (2023)
- [ScienceWorld: Is your Agent Smarter than a 5th Grader?](https://mlsanity.com/paper/2203.07540/) (2022)
- [SWE-RL: Advancing LLM Reasoning via Reinforcement Learning on Open Software Evolution](https://mlsanity.com/paper/2502.18449/) (2025)
- [Can Large Language Models Infer Causation from Correlation?](https://mlsanity.com/paper/2306.05836/) (2023)
- [CLadder: Assessing Causal Reasoning in Language Models](https://mlsanity.com/paper/2312.04350/) (2023)
- [DataSciBench: An LLM Agent Benchmark for Data Science](https://mlsanity.com/paper/2502.13897/) (2025)
- [Process-Verified Reinforcement Learning for Theorem Proving via Lean](https://mlsanity.com/paper/2606.20068/) (2026)

## Abstract (arXiv)

Reinforcement learning with verifiable rewards (RLVR) has advanced language-model reasoning in domains such as mathematics and code, where objective answers are inexpensive to check. Diagnostic reasoning over complex data lacks this advantage: establishing the true cause of an anomaly often requires costly expert investigation and may remain ambiguous after the fact. We ask whether this asymmetry of verification can instead be engineered. We sample an intervention, inject it into a controlled simulator, and generate the observations it would produce. The hidden intervention provides an oracle label and objective reward, while the agent must still investigate noisy, confounded, and distributed evidence. We instantiate this approach in TRACE, a digital-advertising diagnostic environment with 12 root causes and fine-grained segment attribution. Agents investigate each episode using Python and SQL and must identify both the root cause and, when applicable, the affected segment assignment. On a held-out 235-episode test set, the strongest prompted baseline, Claude Opus 5, reaches 0.686 FullAttr@1. Supervised fine-tuning raises Qwen3.5-35B-A3B from 0.159 to 0.637, and subsequent RL with synthesized rewards reaches 0.757, outperforming all evaluated prompted baselines, including frontier closed-source models and a prompted Qwen3.5-122B-A10B model. The resulting policy also uses substantially fewer tool calls than the prompted 35B base. These results provide evidence that access to a scalable, objective training signal can be a more important constraint than model scale alone. More broadly, simulation-based verification can make otherwise ambiguous diagnostic reasoning tasks amenable to scalable reinforcement learning.