Loading
Life sciences · Preprint
arXiv · September 9, 2026
Raises a question worth testing. It does not answer one.
TRACE is a reinforcement-learning framework that uses simulation-based verification to train reasoning agents on diagnostic tasks where ground truth is normally expensive or ambiguous to obtain. In a controlled digital-advertising diagnostic simulator with 12 root causes, a fine-tuned Qwen model trained with synthesized rewards achieved 0.757 FullAttr@1, outperforming several prompted baselines. This is an early-stage technical demonstration that simulation-based objective rewards can improve agent reasoning; it does not establish clinical utility, generalizability beyond the synthetic task, or value for real-world diagnostic reasoning.
Single-arm computational evaluation in simulated diagnostic environment. Computational agents (language models: Claude Opus 5, Qwen3.5-35B-A3B, Qwen3.5-122B-A10B) tested on a synthetic digital-advertising diagnostic task with 12 root causes and segment attribution.. Intervention: Reinforcement learning with synthesized rewards derived from simulation-based verification (oracle labels generated by injecting hidden interventions into a controlled simulator).. Compared with: Prompted baselines (Claude Opus 5, prompted Qwen3.5-122B-A10B) and supervised fine-tuned baseline (Qwen3.5-35B-A3B)..
Strongest prompted baseline (Claude Opus 5) achieved 0.686 FullAttr@1 on held-out 235-episode test set. Supervised fine-tuning of Qwen3.5-35B-A3B improved performance from 0.159 to 0.637. Reinforcement learning with synthesized rewards reached 0.757 FullAttr@1, outperforming all evaluated prompted baselines.
Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.
The source did not state who this applies to in practice.
This is a proof-of-concept study of a machine-learning method (TRACE) in a simulated diagnostic environment; it demonstrates technical feasibility and relative performance gains in an artificial task, but does not address clinical validity, real-world diagnostic accuracy, or applicability to actual clinical reasoning.
As stated by the source record.
Quoted from the source exactly as published.
Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.
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.
Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.