DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training

Hugging Face Daily Papers Papers

Summary

DRACO is a reinforcement learning method that dynamically generates rubrics and redistributes trajectory scores to improve long-horizon agent performance without verifiers.

Reinforcement Learning from Verifiable Rewards works well when a task has a programmatic checker, but most long-horizon agent domains have none. We work in the outcome-blind setting, where ground-truth success signals are not available. Multi-criteria rubrics are a popular way to supply such a reward; they are scored once per trajectory, but a single scalar is a poor signal across tens of steps. We propose DRACO: Distributing Rubric-based Advantage for Credit Optimization. It generates rubrics dynamically during training to track the policy's evolving capability, scores those rubrics once per completed trajectory, and redistributes that judgment over the steps responsible for annotated rubrics to produce differentiated per-step advantages in GRPO. The redistribution is closed-form and does not introduce any trained attribution module. On AppWorld, DRACO gains 15.9 points over the base model and 5.3 points over GRPO trained with a sparse ground-truth reward, despite not using any verifiers itself. On out-of-domain Tau-Bench, it gains 5.3 points over the base model even without a frontier judge, beating both ground-truth-reward training and other rubric-based training settings. The code for DRACO is available at https://github.com/IBM/draco.
Original Article
View Cached Full Text

Cached at: 09/04/26, 03:58 PM

Paper page - DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training

Source: https://huggingface.co/papers/2609.04094

Abstract

DRACO dynamically generates rubrics and redistributes trajectory-level scores into per-step advantages for reinforcement learning without verifiers, improving long-horizon agent performance.

Reinforcement Learning from Verifiable Rewardsworks well when a task has a programmatic checker, but most long-horizon agent domains have none. We work in theoutcome-blindsetting, where ground-truth success signals are not available.Multi-criteria rubricsare a popular way to supply such a reward; they are scored once per trajectory, but a single scalar is a poor signal across tens of steps. We proposeDRACO: DistributingRubric-based AdvantageforCredit Optimization. It generates rubrics dynamically during training to track the policy’s evolving capability, scores those rubrics once per completed trajectory, and redistributes that judgment over the steps responsible for annotated rubrics to produce differentiatedper-step advantagesinGRPO. The redistribution is closed-form and does not introduce any trained attribution module. On AppWorld,DRACOgains 15.9 points over the base model and 5.3 points overGRPOtrained with a sparse ground-truth reward, despite not using any verifiers itself. On out-of-domain Tau-Bench, it gains 5.3 points over the base model even without a frontier judge, beating both ground-truth-reward training and other rubric-based training settings. The code forDRACOis available at https://github.com/IBM/draco.

View arXiv pageView PDFGitHub0Add to collection

Get this paper in your agent:

hf papers read 2609\.04094

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.04094 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.04094 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.04094 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

DACA-GRPO: Denoising-Aware Credit Assignment for Reinforcement Learning in Diffusion Language Models

arXiv cs.LG

This paper identifies weaknesses in existing reinforcement learning methods for diffusion language models—lack of temporal credit assignment and biased likelihood estimates—and proposes DACA-GRPO, a plug-and-play enhancement that introduces denoising progress scores and stratified masking likelihood, achieving consistent improvements across reasoning, code generation, and constrained generation benchmarks.

Not Every Rubric Teaches Equally: Policy-Aware Rubric Rewards for RLVR

Hugging Face Daily Papers

This paper introduces POW3R, a policy-aware rubric reward framework for reinforcement learning with verifiable rewards (RLVR). It shows that static rubric aggregation misallocates learning signal, and POW3R achieves faster convergence and better performance across multiple settings.