AI数据分析系统自动评分：严格匹配与宽松评估的权衡

Skip to main content

arXiv is now an independent nonprofit!Learn more×

Search arXiv

Press Enter to search ·Advanced search

Computer Science > Artificial Intelligence

arXiv:2606.24839v1 (cs)

[Submitted on 23 Jun 2026]

Title:Grading the Grader: Lessons from Evaluating an Agentic Data Analysis System

Authors:Tian Zheng,Kai-Tai Hsu

View a PDF of the paper titled Grading the Grader: Lessons from Evaluating an Agentic Data Analysis System, by Tian Zheng and Kai-Tai Hsu

View PDFHTML (experimental)

Abstract:Agentic data analysis systems produce rich outputs, including code, numerical results, and verbal diagnostics. This makes them more challenging to evaluate than single-turn LLM responses. It is therefore necessary to distinguish genuine disagreement between an agent's output and a ground-truth answer from grading artifacts. We investigate how reliably automated graders assess such a system and what strategies improve grading quality by applying LAMBDA, a multi-agent data-analysis system, on 153 numerical QRData tasks from DSGym. We develop and evaluate a three-layer human-AI grading cascade: strict regex matching, LLM-based lenient grading, and snippet-based human inspection, which combines non-GenAI and GenAI strategies with different failure profiles. Both automated graders achieve 100% observed precision (0/70 false positives). The lenient grader's recall is 97% against human labels. A keyword-anchored extraction pipeline raises the strict grader's recall by 60 percentage points over a last-number heuristic; the lenient grader is architecturally parser-independent. An iterative nudge mechanism raises grading run success from 36% to 97% and lenient-pass rates from 16% to 46%; comparing nudging with and without original-question re-injection shows that re-injection offers no benefit, confirming the nudge as an answer template cue. We further observe in this case study that variable type is the task metadata field most consistently associated with grading pipeline dynamics and observed outcome grades.

| | | | --- | --- | | Subjects: | Artificial Intelligence (cs.AI); Applications (stat.AP) | | MSC classes: | 68T42 | | ACM classes: | I.2.1 | | Cite as: |arXiv:2606.24839[cs.AI] | | | (orarXiv:2606.24839v1[cs.AI] for this version) | | |https://doi.org/10.48550/arXiv.2606.24839<br>Focus to learn more<br>arXiv-issued DOI via DataCite |

Submission history

From: Tian Zheng \[view email]

[v1] Tue, 23 Jun 2026 17:18:28 UTC (678 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Grading the Grader: Lessons from Evaluating an Agentic Data Analysis System, by Tian Zheng and Kai-Tai Hsu

• View PDF
• HTML (experimental)
• TeX Source


Current browse context:

cs.AI

[< prev](https://arxiv.org/prevnext?id=2606.24839&function=prev&context=cs.AI "previous in cs.AI (accesskey p)")  \|  [next >](https://arxiv.org/prevnext?id=2606.24839&function=next&context=cs.AI "next in cs.AI (accesskey n)")

new\|recent\|2026-06

Change to browse by:

cs

stat

stat.AP

References & Citations

• NASA ADS
• Google Scholar
• Semantic Scholar

export BibTeX citation

Bookmark


Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer _(What is the Explorer?)_

Connected Papers Toggle

Connected Papers _(What is Connected Papers?)_

Litmaps Toggle

Litmaps _(What is Litmaps?)_

scite.ai Toggle

scite Smart Citations _(What are Smart Citations?)_

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv _(What is alphaXiv?)_

Links to Code Toggle

CatalyzeX Code Finder for Papers _(What is CatalyzeX?)_

DagsHub Toggle

DagsHub _(What is DagsHub?)_

GotitPub Toggle

Gotit.pub _(What is GotitPub?)_

Huggingface Toggle

Hugging Face _(What is Huggingface?)_

ScienceCast Toggle

ScienceCast _(What is ScienceCast?)_

Demos

Demos

Replicate Toggle

Replicate _(What is Replicate?)_

Spaces Toggle

Hugging Face Spaces _(What is Spaces?)_

Spaces Toggle

TXYZ.AI _(What is TXYZ.AI?)_

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower _(What are Influence Flowers?)_

Core recommender toggle

CORE Recommender _(What is CORE?)_

• Author
• Venue
• Institution
• Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community?Learn more about arXivLabs.

Which authors of this paper are endorsers?\| Disable MathJax (What is MathJax?)