AdaPRA: Process Reward Agents with Adaptive Rubrics for Data Science

Hugging Face Daily Papers Papers
data-science-agents process-reward-model rubric-guided-supervision llm-agents machine-learning-workflows evaluation

Summary

AdaPRA introduces a persistent coding agent harness with adaptive rubric-guided process supervision that accumulates knowledge across attempts, boosting an 80B Qwen model's leaderboard percentile by 24.5 points on DSPredict-Easy and 14.8 on DSPredict-Hard while reducing process errors in data science agents.

Data science agents can automate end-to-end machine learning workflows, yet they remain far from competitive on human leaderboards. Agents make process errors such as insufficient exploration, unsound validation, coding bugs, and poor use of a fixed budget (re-running the same exploratory data analysis and baseline models in every attempt). We propose AdaPRA to address these issues: a persistent coding agent harness that accumulates knowledge across attempts on the same task, with rubric-guided process supervision that identifies and corrects errors. Our process reward agent audits the agent's trajectory and workspace by executing checks, and adapts its rubrics to each task. We find that rubrics seeded from easier tasks generalize to harder tasks, with rubric adaptation further improving performance. With five attempts per task, AdaPRA improves an 80B Qwen model's leaderboard percentile by 24.5 points on DSPredict-Easy and 14.8 points on DSPredict-Hard while reducing process errors. We also show that supervision with rubrics mined from one base model's runs improves another (DeepSeek-V3.2), and that AdaPRA beats DataPRM, a competitive PRM for data science. We also share our code for reproducibility.
Original Article
View Cached Full Text

Cached at: 10/29/26, 01:32 AM

Paper page - AdaPRA: Process Reward Agents with Adaptive Rubrics for Data Science

Source: https://huggingface.co/papers/2610.33245

Abstract

AdaPRA improves data science agents via persistent knowledge accumulation and adaptive rubric-guided process supervision, reducing errors and boosting leaderboard performance.

Data science agentscan automate end-to-endmachine learning workflows, yet they remain far from competitive on human leaderboards. Agents makeprocess errorssuch as insufficient exploration, unsound validation, coding bugs, and poor use of a fixed budget (re-running the same exploratory data analysis and baseline models in every attempt). We propose AdaPRA to address these issues: a persistent coding agent harness that accumulates knowledge across attempts on the same task, with rubric-guidedprocess supervisionthat identifies and corrects errors. Ourprocess reward agentaudits the agent’s trajectory and workspace by executing checks, and adapts its rubrics to each task. We find that rubrics seeded from easier tasks generalize to harder tasks, withrubric adaptationfurther improving performance. With five attempts per task, AdaPRA improves an 80B Qwen model’sleaderboard percentileby 24.5 points on DSPredict-Easy and 14.8 points on DSPredict-Hard while reducingprocess errors. We also show that supervision with rubrics mined from one base model’s runs improves another (DeepSeek-V3.2), and that AdaPRA beatsDataPRM, a competitive PRM for data science. We also share our code for reproducibility.

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2610\.33245

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/2610.33245 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.33245 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.33245 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

AdaPRA: Process Reward Agents with Adaptive Rubrics for Data Science

arXiv cs.AI

AdaPRA 提出了一种用于数据科学智能体的持久化编码框架与过程奖励智能体（PRA），通过自适应评分标准（rubrics）检测并纠正过程错误，使 80B Qwen 模型在 DSPredict 基准上的排名百分位提升了 24.5（Easy）和 14.8（Hard）个点。

ARBOR: Online Process Rewards via a Reusable Rubric Buffer for Search Agents

arXiv cs.CL

ARBOR introduces a reusable rubric buffer to provide online process rewards for LLM-based search agents, improving training efficiency when outcome-only rewards are insufficient. It outperforms GRPO and DAPO on multi-hop QA benchmarks, converting up to 42% of zero-gradient training groups into informative ones.

Prime Agent: A Self-Improving RLM Harness

Hugging Face Daily Papers

Prime Agent is an open-source harness that uses recursive subagents and persistent computation to extend language models' long-horizon capabilities across coding and reasoning tasks, significantly improving performance on benchmarks like ARC-AGI-3.

PRO-CUA: Process-Reward Optimization for Computer Use Agents

arXiv cs.AI

This paper introduces PRO-CUA, a process-reward optimization framework for training Computer Use Agents (CUAs) using iterative step-level reinforcement learning. The method decouples on-policy environment interaction from policy optimization, enabling dense credit assignment without relying on expert trajectories, and demonstrates effectiveness on live web benchmarks.