arXiv Science⌕ Search

arXiv · 2611.23250

What Moves the Ceiling on Automated Audit: from General Code to Scientific Code, Results and Data

Abstract

Models increasingly check what other models write. We study one shipped arrangement, in which an auditor from a different vendor reads the work and deterministic checks verify what code can verify, and measure its limits against ground truth no model supplies. On general-purpose code with hidden tests, eight readings of the auditor reach 30.0% recall [20.0, 40.7] at 16.0% [10.1, 22.3] flags on correct code, with diminishing returns and no plateau. The reading model (confounded with its sampling), one rulebook sentence and the decision rule each move that figure, and much of what is missed is, on a post hoc rating, behaviour the specification never determined. In three preregistered science studies, checking that reported numbers trace to their sources flagged all 49 inconsistent reports and none of 25 consistent fabrications, of which the model auditor flagged 23 (21 with a sound finding, post hoc); on data, a documentation-derived validator and the auditor flagged overlapping but different faulty items, together 112 of 140 at a 2.9% [0.7, 5.7] flag rate on clean data; and on scientific code, reframing the task around one step cut flags on correct code from 68.7% to 36.0%. Where both tiers ran, the deterministic one flagged no clean item and could not see consistent error; the model saw some and erred itself. Studies making model calls were preregistered, the two unregistered analyses are named, and every quoted report passed independent cross-vendor review.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Zhaohe Dong. 2026-09-24. What Moves the Ceiling on Automated Audit: from General Code to Scientific Code, Results and Data. https://arxiv.science/abs/2611.23250

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

AgentLens: Revealing The Lucky Pass Problem in SWE-Agent Evaluation

Evaluation of software engineering (SWE) agents is dominated by a binary signal: whether the final patch passes the tests. This outcome-only view treats a principled solution and a chaotic trial-and-error process as equivalent. We show that this equivalence is empirically false. We evaluate 2,614 OpenHands trajectories from eight model backends on 60 SWE-bench Verified tasks. Of these, 47 have enough passing trajectories to construct task-level process references, yielding a 1,815-trajectory evaluation subset. Among passing trajectories in this subset, 10.7% exhibit behavior we call a Lucky Pass: regression cycles, blind retries, missing verification, or temporally disordered exploration, implementation, and verification. We introduce AgentLens, a framework for process-level assessment of SWE-agent trajectories, and release AgentLens-Bench, a dataset of 1,815 trajectories annotated with quality scores, waste signals, divergence points, and 47 task-level Prefix Tree Acceptor (PTA) references. AgentLens builds PTA references by merging multiple passing solutions for the same task, and uses a context-sensitive intent labeler to assign actions to Exploration, Implementation, Verification, or Orchestration based on trajectory history rather than tool identity alone. On AgentLens-Bench, the quality score separates passing trajectories into Lucky, Solid, and Ideal tiers and further decomposes Lucky Passes into five recurring mechanisms. Across the eight model backends, Lucky rates range from 0.5% to 23.2%, and some models move by as many as five rank positions when ranked by quality score instead of pass rate. We publicly release AgentLens-Bench and the AgentLens SDK at https://github.com/agentlens-framework/agentlens, with a project page at https://agentlens-framework.github.io/agentlens/.

cs.SE↗

Where Is the Cost of Third-Party API Routers in Agentic Software Development?

Third-party API routers have become a common layer that unifies access across increasingly diverse LLM providers. In coding-agent workflows, high-autonomy operation is widely adopted because it reduces interaction overhead. As a result, a third-party API router, which sits between the agent and the upstream provider, inevitably occupies the trusted path. It can inspect and modify every request and response, yet no mechanism verifies alignment between the provider's output and the repository-level actions ultimately executed by the agent. Consequently, client-side permission mechanisms may become ineffective in practice. Whether this control gap produces real, hard-to-detect effects on software development tasks remains empirically unmeasured. In this paper, we conduct an empirical study of router-side injection in coding agents, examining four intervention levels of increasing subtlety: Response Substitution (L1), Response Append (L2), LLM-Polished Injection (L3), and LLM-Polished with Distribution Alignment Injection (L4). Moreover, we develop SIDEL, a framework for trace recording, replay, injection, and defense evaluation, with a curated dataset of 400 samples. We evaluate four representative coding agents, and further evaluate whitelist-based execution control and LLM review. Router-side intervention substantially alters repository-level actions and remains difficult for existing client-side safeguards to detect. Without additional mitigations, all evaluated agents achieved a defense success rate of 0 percent across all injection levels. Client-side mitigations and reactive reviews improve resistance but do not fully restore end-to-end control, motivating provider-side output-integrity guarantees. Our code is available at https://github.com/Riyasushin/SIDEL.

cs.SE↗

A Grey Literature Review of Graduate Students' Research Practices and Skills in the LLM Era

Context: Graduate students in computing now use Large Language Models (LLMs) throughout their research, from literature reviews and coding to data analysis and writing. Yet there is little shared understanding of which skills competent use of these tools requires, so graduate programs struggle to teach, mentor, or assess it, and unverified outputs can propagate into the scientific record. Objective: We investigate which LLM-based practices graduate students employ in their research and which skills they need to benefit from these practices. Method: We conducted a grey literature review of Reddit discussions. We retrieved 1,383 posts from five research-oriented subreddits, screened them against explicit inclusion and exclusion criteria, and analyzed the 54 included posts with thematic analysis, in which two authors independently coded each excerpt. Results: We identified seven categories of practices and nine categories of skills. Commenters locate the problem in accepting generated output without inspection rather than in using LLMs, and warn that delegating the effortful parts of a task, such as debugging, reading, or validating sources, delivers the output without building the skill the task was meant to develop. Traditional skills, such as programming and critical thinking, are what allow students to verify what an LLM produces. Conclusions: Our categories give graduate programs a concrete basis to teach students to work with LLMs without eroding the skills that research training should build, and they add research-specific skills, such as data sharing and scientific reproducibility, to existing AI competency frameworks.

cs.SE↗