https://arxiv.org/api/nRuDmKfp2vHvzwBt7fhAi/87vjM 2026-09-06T21:27:39Z 300 55 0 http://arxiv.org/abs/2609.04167v1 2026-09-03T17:53:34Z

Repository-level software engineering benchmarks have significantly advanced the evaluation of coding agents, but existing benchmarks primarily measure whether generated patches pass functional tests and overlook review-derived acceptance constraints (review constraints) that often influence whether a patch is acceptable in real-world software development. We introduce SWE-Gate, a repository-level benchmark for software engineering agents that explicitly evaluates review constraint compliance alongside functional correctness. SWE-Gate derives review constraints from real pull request review comments and synthesizes repository-level repair instances around these constraints. Each instance provides separate functional and constraint tests, together with non-compliant and gold patches, enabling explicit separation between issue resolution capability and review constraint compliance. We construct SWE-Gate with 303 repository-level repair instances spanning 75 open-source Python repositories across diverse software domains. Experiments with four LLM backends spanning different capability levels under a common coding-agent scaffold reveal a substantial gap between functional success and success under the complete repair specification: among 644 repairs that pass the functional tests, 221 fail to satisfy the provided review constraints. These findings show that functional-only evaluation overestimates agents' ability to satisfy the full requirements of repository-level repair tasks. The replication package including code, data, and experimental results is available at https://github.com/DeepSoftwareAnalytics/SWE-Gate.

2026-09-03T17:53:34Z 11 pages, 2 figures, 5 tables Xin He Yanlin Wang Mingwei Liu Jiachi Chen Hongyu Zhang Guanbin Li http://arxiv.org/abs/2609.04123v1 2026-09-03T17:24:15Z

Checking a manuscript's reference integrity, its compliance with a target venue's specific submission rules, and its adherence to community reporting standards is manual, repetitive, and different for every venue so in practice it is done inconsistently or skipped. We present ATIBA, a tool that runs five grounded integrity and quality checks on a manuscript: a reference-integrity check that verifies each citation against bibliographic sources and flags retracted or unfindable references; a venue/track compliance check that derives submission criteria directly from a venue's own call-for-papers page and evaluates the manuscript against them, each verdict anchored to a verbatim quote from that page; an empirical-standards compliance check against the ACM SIGSOFT Empirical Standards, with a hallucination defence that discards any evidence quote it cannot locate verbatim in the manuscript; a multi-mode AI review (venue-specific, formal, and page-anchored annotation) powered by GPT-5.4 through Azure OpenAI; and a citation-suggestion feature that proposes candidate references for a manuscript and verifies each against bibliographic sources before it is shown to the user. All five checks are designed around the same principle: an LLM is only trusted to judge, never to invent the evidence it judges against. We evaluated ATIBA through a moderated user study with 13 non-author participants. Agreement across the six survey items ranged from 69% to 92%, with a mean of 85%, providing initial evidence of positive perceived usefulness across the evaluated workflows. These findings establish perceived usefulness; objective accuracy remains to be measured.

2026-09-03T17:24:15Z Accepted at The First International Workshop on Automated Techniques for Integrity and Quality in Software-Engineering Research (ATIQSER 2026) Veli Karakaya Semih Çağlar Yusuf Yiğit Korkmaz Eray Tüzün http://arxiv.org/abs/2609.04094v1 2026-09-03T17:02:20Z

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.

2026-09-03T17:02:20Z Shubham Gandhi Saurabh Goyal Kiran Kate Yara Rizk http://arxiv.org/abs/2609.04075v1 2026-09-03T16:44:22Z

AI agents have recently demonstrated strong performance in automated vulnerability patching. However, existing evaluations often validate a patch only by testing whether the provided Proof-of-Concept (PoC) input still triggers a crash. This leaves two key threats to validity: agents may reproduce memorized historical developer patches, or they may generate surface-level fixes that only suppress the reported crash. We study these concerns for C/C++ vulnerability patching. We introduce a patch similarity metric to detect memorized patches. On average, 25% of the agent patches exhibit substantial similarity to historical developer patches, indicating that patch memorization is a real threat to the validity of vulnerability patching evaluations. Meanwhile, agents also frequently exploit benchmark structures to pass patch validation by patching on the crash stack trace to suppress the crash, rather than localizing and fixing the root cause of the vulnerabilities. To handle these issues, we propose PatchBench, a new benchmark for evaluating AI agents on realistic vulnerability patching tasks. PatchBench selects vulnerabilities whose ground-truth fixes lie outside the crash stack and uses vulnerability transplant and code mutations to migrate historical vulnerabilities into new repository contexts, reducing the risks of surface-level fixes and patch memorization. We develop new patch validation methods that thoroughly evaluate both security and semantic correctness of agent patches. Across 11 state-of-the-art agents, including the top three AIxCC agents, the original PoC-only validation inflates the patching task solve rate of agents by 1.83$\times$ on average. Our results reveal key limitations of current patching agents and point to future research directions for more reliable vulnerability repair.

2026-09-03T16:44:22Z Chihao Shen Jiacheng Li Aastha Mahajan Jeffery Siyuan Tian Yonghwi Kwon Yizheng Chen http://arxiv.org/abs/2609.04061v1 2026-09-03T16:36:05Z

Large language models (LLMs) are increasingly used to edit existing code, but correctness alone is not enough: useful repairs should also be minimal, reviewable, and faithful to the original implementation. We study over-editing, the tendency of a model to rewrite code beyond what is required to fix a bug. We construct an evaluation framework from 400 BigCodeBench problems by injecting controlled AST-level corruptions into reference solutions, giving each repair task a known minimal patch. Across frontier LLMs, over-editing is widespread even among strong models like GPT-5.5: high Pass@1 can coexist with unnecessarily large edits and added cognitive complexity. A preservation instruction substantially reduces this behavior, lowering average excess Levenshtein distance from 0.195 to 0.131, reducing added cognitive complexity by 26.6%, and increasing Pass@1 by 2.3 points. However, these gains do not simply follow from a larger reasoning budget or larger models. We next ask whether minimal editing can be learned directly during post-training. We observe that supervised fine-tuning overfits to seen corruption patterns, whereas reinforcement learning gives the best out-of-domain edit-fidelity and performance-retention trade-off. These results position edit fidelity as a distinct axis of code-repair quality and show that it can be measured and learned.

2026-09-03T16:36:05Z EMNLP 2026 (Main) Tongyao Zhu Wei Hern Lim Min-Yen Kan http://arxiv.org/abs/2609.04055v1 2026-09-03T16:31:42Z

Software users often submit issue reports to a product's issue tracking system to report defects, suggest enhancements, or raise other product-related concerns. Labeling these issue reports supports effective planning and improves community engagement. However, many issue reports remain unlabeled due to the substantial manual effort required to design an appropriate label taxonomy, then assign suitable labels from this taxonomy to new issue reports. Existing automated labeling approaches attempt to mitigate these challenges. However, they suffer from key limitations, such as extensive manual intervention, the assignment of generic labels, and a dependence on existing labeled datasets. To address these limitations, we propose LabelMate, a novel Large Language Model (LLM)-driven framework that (1) derives a comprehensive, project-specific label set from historical issue reports and (2) automatically assigns relevant labels to new issue reports without requiring any pre-labeled training data. We evaluate LabelMate on 16,500 issue reports from 30 popular and diverse GitHub repositories. Based on this dataset, our approach generates a coherent list of 275 labels and achieves an average labeling accuracy of 89.84%, a statistically significant improvement over existing generic label assigning approaches. These results demonstrate that LabelMate offers an efficient, domain-adaptive solution to streamline the issue labeling process.

2026-09-03T16:31:42Z 38 pages, 11 figures Liam Johnston Shayan Noei Maram Assi Ying Zou http://arxiv.org/abs/2609.03894v1 2026-09-03T14:12:42Z

Large language models (LLMs) have become ubiquitous tools for code generation and editing. However, development teams often use multiple LLM assistants. Different developers may prefer different models, and individual developers may switch between models across different coding sessions. Because of this, the edits any one model makes are frequently applied to foreign code originally generated by another model. These LLMs are often trained on different datasets, and as a result have different stylistic preferences. Do LLMs behave differently when they edit foreign code originally written by a different LLM with a different coding style? We find that models tend to make more, and often excessive, edits on foreign code. We introduce CROCODIL (Cross-model Code Editing with LLMs), a post-training framework for reducing excessive edits while preserving functional correctness. CROCODIL's similarity reward penalizes large changes, while its execution reward scores build and test success. We use the product of these two rewards to encourage the policy to decrease the edit size without decreasing the edit task success rate. CROCODIL is available at https://github.com/EngineeringSoftware/Crocodil.

2026-09-03T14:12:42Z EMNLP 2026 Findings Linghan Zhong Aditya Thimmaiah Jayanth Srinivasa Milos Gligoric Junyi Jessy Li http://arxiv.org/abs/2609.03861v1 2026-09-03T13:50:39Z

Software engineers expect a software system to be efficient, maintainable, economically viable, and socially responsible throughout its lifecycle. Unfortunately, decisions made at the organizational, process, or product levels of software development - even when they improve one of these goals - often have unintended long-term consequences at other levels. In response, software-engineering research has explored different ways to achieve software sustainability, including energy efficiency, resource optimization, and code maintainability. However, existing research offers limited explanations for how sustainability problems emerge and reinforce one another across the socio-technical levels of software engineering. Moreover, we argue that sustainability challenges are not isolated but rather systemic: They stem from interactions among organizational priorities, development practices, and technical conditions. In this vision paper, we introduce the concept of sustainability antipatterns to capture conditions that systematically produce unsustainable outcomes. We demonstrate how this concept can help expose cross-level misalignments between socio-technical levels and sustainability dimensions that remain difficult to recognize through artifact-centric or dimension-specific analyses. To guide future research, we outline concrete directions for identifying, formalizing, detecting, and mitigating sustainability anti-patterns. We envision sustainability anti-patterns as a foundation for an integrated and actionable research agenda on sustainable software engineering.

2026-09-03T13:50:39Z Accepted for publication at the 2026 IEEE International Conference on Software Maintenance and Evolution (ICSME) Masoum Salehi Sandro Schulze Jacob Krüger http://arxiv.org/abs/2609.03778v1 2026-09-03T12:51:10Z

In this paper, we introduce a notion of abstract quantum data type within the framework of universal algebra. This notion provides an algebraic foundation for describing data abstraction in quantum programming. We formally define a quantisation of classical data types and show that their equational specifications can be soundly lifted to the quantum setting. Two standard quantisation methods for classical functions, namely the bit oracle and the phase oracle, arise as special cases of this general construction. We illustrate the framework with applications to quantum arrays and quantum error-correcting codes, showing how they can be understood through the lens of data-type quantisation. We further establish conditions under which quantisation preserves structural relationships and constructions of classical data types, including embeddings, isomorphisms, and products.

2026-09-03T12:51:10Z 37 pages Mingsheng Ying Zhicheng Zhang Kean Chen http://arxiv.org/abs/2609.03760v1 2026-09-03T12:28:19Z

Simulation is increasingly used to support safety-related decision-making in road transport, particularly for the assessment and approval of automated driving systems (ADS). The complexity of ADS behavior and size of their operational design domains make exclusive reliance on physical testing impractical, leading to extensive use of virtual testing (VT) during the approval phase. This shift raises critical questions regarding the credibility of modelling and simulation (M&S) results used to support road safety decisions. Current VT accreditation approaches in the ADS domain typically rely on validation-only practices, which have been shown to scale poorly when applied to complex, multi-tool simulation environments. To address this limitation, this paper proposes a risk-based framework for assessing the credibility of simulation toolchains used in ADS safety evaluation, drawing inspiration from established practices in other safety-critical domains, notably NASA's STD-7009 for models and simulations. The framework extends traditional verification and validation (V&V) by explicitly linking credibility requirements to the intended use of simulation outputs and to the safety criticality of the decisions they support within the approval process. It provides a lifecycle-oriented assessment scheme integrating toolchain management, modelling assumptions and limitations, verification, validation, and sensitivity analysis. Credibility acceptance thresholds are defined proportionally, allowing differentiated requirements depending on whether simulation is used for exploratory safety analysis, partial decision support, or as a substitute for physical testing. While demonstrated for ADS, the proposed approach is directly applicable to road safety and simulation studies where VT plays a central role in safety assessment and regulatory decision-making.

2026-09-03T12:28:19Z Road Safety and Simulation 2026 RSS2026 Riccardo Dona Espedito Rusciano Biagio Ciuffo http://arxiv.org/abs/2609.03721v1 2026-09-03T11:54:27Z

Context: Architectural Design Decisions (ADDs) capture the rationale behind the structure and evolution of software systems but are rarely documented explicitly, and are often hidden inside source code commits. Recovering them is important for Architectural Knowledge Management (AKM). Problem: Extracting ADDs from commits is challenging due to their implicit and unstructured nature. Large Language Models (LLMs) have shown strong capabilities in understanding code and text, yet their effectiveness for this task remains underexplored. Study: We present a preliminary study using four LLMs (Gemini 3 Pro, DeepSeek R1, Kimi K2, Qwen3) with zeroshot and fewshot prompting on 30 developer-written ADDs from open-source projects. We score outputs with ROUGE-L, BLEU, METEOR, and BERTScore, and one author manually reviews the Gemini outputs. Results: All models reach a BERT-F1 above 0.81, and fewshot prompting improves alignment (Gemini BERT-F1: 0.828 to 0.847). However, the generated ADDs are often too long, implementation-focused, and miss the rationale behind the decision. This highlights opportunities for architecture-aware LLM systems and automated AKM.

2026-09-03T11:54:27Z Accepted at IdeaArch Workshop at ECSA 2026 Amey Karan Rudra Dhar Mohamed Soliman Karthik Vaidhyanathan http://arxiv.org/abs/2609.03592v1 2026-09-03T09:40:30Z

Due to their black-box nature, LLMs suffer from limited explain- ability and a lack of determinism. Their usage cost can also rise, particularly with repetitive tasks on large codebases. To mitigate this, we conduct a novel empirical study targeting three domain- specific languages for transformation rules, namely Comby, GritQL, and Ast-Grep. We evaluate three LLMs (GPT-5.4, GPT-oss-120B, and Llama3.1-8B) on six diverse datasets covering four software- evolution tasks: API misuse correction, program repair, API migra- tion, and language version migration. Our results provide evidence that transformation rule synthesis moves beyond proof-of-concept with strong frontier models. GPT-5.4 achieves consistently high rule applicability rates and produces transformations closest to the ground truth across most benchmarks. Smaller and open-weight GPT-oss-120B and Llama3.1-8B models remain effective for simpler, localized changes but struggle with complex migration scenarios. We also observe non-negligible generalizability through the usage of meta-variables and through a high reuse score in the first quartile of many datasets. Finally, when compared to the anti-unification algorithm, LLMs outperform it in correctness, but underperform in rule applicability. Overall, our results show great potential for LLMs to generate sound, correct, generalizable, and reusable rules.

2026-09-03T09:40:30Z Accepted at the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE 2026) Axel Allain Aymeric Blot Djamel Eddine Khelladi Mathieu Acher 10.1145/3832783.3837504 http://arxiv.org/abs/2609.03456v1 2026-09-03T07:15:05Z

Artificial intelligence (AI) is increasingly used to augment software engineering (SE) workflows. While code generation remains the main use case, organizations are actively seeking AI integration in other practices such as test cases generation and code reviews. Organizational AI adoption strategies seem to focus on tangible outcomes such as productivity. However, AI is a disruptive force, introduced into settings where role identity, team norms, and the sources of job satisfaction were well established before the recent advances in generative AI. Historically, technological disruptions have caused psychological and social strains in workplaces, ranging from anxiety and eroded meaning to deskilling and disrupted professional identities. The assumption that AI for SE is cost-free may not be accurate. Therefore, in this study we sought to understand the psychological costs software professionals experience during organizational AI adoption. We carried out a case study in a large software development services company, one year after the company launched its AI adoption. We collected qualitative data through meetings and semi-structured interviews (N = 21). We found that software professionals experience accountability anxiety, craft identity disruption, meaning and satisfaction erosion, cognitive and workload intensification, and uncertainty distress. Practitioners manage these costs through practices that restore control, mitigate them through protective and identity-preserving adaptations, or absorb them, carrying what neither can resolve. We contribute to AI-human collaboration in SE by repositioning AI adoption as a human transition, not only a technological and organizational one.

2026-09-03T07:15:05Z Adam Alami Elda Paja Abhishek Tiwari http://arxiv.org/abs/2609.03309v1 2026-09-03T02:58:24Z

Test-time scaling for code generation typically explores the solution space by sampling multiple programs from a fixed instruction. We study a complementary direction: instance-level instruction-space exploration. Our observation is that many coding failures stem from missing constraints, overlooked edge cases, or misleading reasoning paths induced by the original prompt. To address this, we propose TipCoder, a test-time instruction proposer that generates problem-specific auxiliary tips before code synthesis. TipCoder distills multi-turn debugging trajectories into proactive guidance and further optimizes the Proposer with reinforcement learning using a marginal-utility reward. At inference time, it generates both a base solution and a tip-guided solution, and applies a Reward Model for post-hoc selection. This exploration-selection design allows tips to expose additional candidate potential while reducing regressions from unnecessary guidance. Across the evaluated code-generation benchmarks and target Code LLMs, TipCoder provides a consistent instruction-level test-time scaling strategy, comparing favorably with stochastic sampling and generic prompt optimization baselines under a shared reward-model-based selection protocol.

2026-09-03T02:58:24Z 15 pages, accepted by findings of EMNLP 2026 Minyu Chen Sihao Wu Ling-I Wu Song Qin Jingyang Li Lei Ning Jianxin Xue Guoqiang Li http://arxiv.org/abs/2609.03267v1 2026-09-03T01:50:06Z

Large language models (LLMs) often produce code that looks plausible but is not grounded in reality. The code may import packages that do not exist or claim to implement algorithms that violate proven theorems, while still compiling and running. We study \emph{code hallucination} as \emph{ungrounded generation} and separate it from ordinary \emph{code error} (bugs in otherwise grounded programs). We propose a taxonomy with three dimensions: \textbf{groundedness} (absolute violations of universal truths vs.\ relative fabrications of contingent or ecosystem-specific facts), \textbf{manifestation level} (syntactic, semantic, or factual), and \textbf{behavior} (from confident fabrication to degenerate output), organized into a severity ordering. We build an \textbf{adversarial} suite of deliberately unsatisfiable tasks where the correct response is to refuse and categorize the responses under our taxonomy. The suite contains \textbf{270 prompts} across six languages and 24 subcategories, paired with \textbf{91 matched solvable controls}, and responses are judged by a two-tier protocol validated against human labels (82\% agreement, $κ{=}0.73$). Across twelve open-weight code and reasoning models (4{,}332 judged responses), models produce ungrounded code on about 60\% of unsatisfiable prompts and refuse only 27\%, while wrongly refusing 0\% of the solvable controls.

2026-09-03T01:50:06Z Vishnu Asutosh Dasu Ashish Kundu Gang Tan http://arxiv.org/abs/2609.03230v1 2026-09-03T00:06:04Z

Requirements engineering (RE) governs the quality of everything downstream in systems engineering (SE); defective requirements that survive review cycles propagate into design rework, schedule delays, and cost overruns. Because requirements are often written in natural language, recent advances in generative AI have raised expectations that large language models (LLMs) can absorb requirement quality assessment, a task otherwise slow and human expertise-intensive. Yet empirical evidence on whether LLMs can be trusted to do so remains scarce. This study presents the first benchmarking analysis of off-the-shelf LLM performance for requirement quality evaluation. Against an expert-derived ground truth built on INCOSE quality criteria, we evaluate ten models spanning two families (OpenAI and Anthropic) and five generations each, across one hundred independent runs, two requirement sets, and five sampling temperatures. Four contributions follow. First, we quantify a strongly asymmetric error profile: across all models and runs, the best-performing Anthropic model detects a median of only 47% of expert-identified issues while false-flagging 11%. Second, performance degrades significantly where SE judgment is required, as necessity and correctness issues are almost always missed. Third, generational progress is non-monotonic, so newer models cannot be assumed better. Fourth, this error behavior shifts only modestly and non-monotonically across sampling temperatures, indicating characteristic model deficiencies rather than inherent stochasticity. Off-the-shelf LLMs are therefore not yet trustworthy autonomous evaluators. Findings also warrant caution for Agentic AI developers: orchestrating these LLM modules in specialized architectures risks compounding these deficiencies rather than correcting them. Their defensible near-term role is human-in-the-loop decision support.

2026-09-03T00:06:04Z Jannatul Shefa Alejandro Salado Paul Wach Taylan G. Topcu http://arxiv.org/abs/2609.03156v1 2026-09-02T20:40:54Z

Large language models (LLMs) are increasingly used in software engineering pipelines for code generation, where production prompts often combine multiple constraints. This paper presents a full-factorial empirical study of how output formatting, persona assignment, and urgency framing jointly affect LLM code-generation reliability. We evaluate all 27 combinations in a controlled 3x3x3 design and decompose each compound condition into an additive prediction and a residual interaction term that captures super-additive degradation. The study uses all 164 HumanEval+ problems across five OpenAI models from the GPT-4o family, GPT-4.1 family, and o3-mini, yielding 22,140 greedy-decoding evaluations. A format-aware extraction pipeline separates formatting failures from reasoning failures, and significance is assessed with McNemar's test, odds ratios, and 95% confidence intervals. Results show that compound constraints can produce architecture-dependent degradation not predictable from single-factor experiments. The GPT-4o family exhibits consistent super-additive effects, with pass@1 reductions 3-12 percentage points beyond additive predictions; the largest interaction is -12.2 pp on GPT-4o-mini for JSON + expert persona + moderate urgency. JSON combinations generally produce larger interactions than XML. In contrast, the GPT-4.1 family is largely resistant, while o3-mini shows a qualitatively different pattern in which structured output constraints can improve performance. These findings show that vulnerability is architecture-dependent rather than size-dependent, that individually neutral or beneficial constraints can combine to cause substantial degradation, and that compound-prompt testing should be standard in reliability assessment for LLM-assisted engineering pipelines.

2026-09-02T20:40:54Z Shrenik Jadhav Nickalsa LaPlaca Caleb Stone Ashok Raja Omar Ochoa Vidhyashree Nagaraju http://arxiv.org/abs/2609.03086v1 2026-09-02T18:59:40Z

This article introduces Eiffel-tools, a language server protocol (LSP) implementation for the Eiffel programming language that uses Large Language Models (LLMs) to aid the development of statically verified software. The tool provides various interactive and non-interactive commands to produce code and specifications. It uses language and project specific knowledge to precisely direct the LLM and verifies the output using a static verifier. It crafts rich programmatic prompts for the input and corrects or rejects the output. Furthermore, it handles the retries until the program passes verification. The tool's bug fixing capability is evaluated on 2 public datasets using 3 models. The tool can fix 76% to 95% of bugs by combining LLMs and a formal verifier depending on the model and prompts used. The results show the trade-off between the number of fixing attempts and the success rate.

2026-09-02T18:59:40Z Submitted and presented at VERIFAI-2026 Alessandro Schena Ilgiz Mustafin Julia Kotovich http://arxiv.org/abs/2609.03028v1 2026-09-02T18:02:18Z

Coding agents often implement changes before users have fully articulated their requirements, echoing a pattern from requirements engineering: stakeholders cannot express a constraint until part of the system exists to react to. This volatility is associated with schedule and budget overruns in traditional projects, but only at release-cycle granularity. Existing work on coding agents narrows this gap only partway: curated benchmarks fix requirements before implementation by design, and observational studies report pushback frequency without linking arrivals to the code invalidation they cause. We address this using 3,553 eligible SWE-chat sessions, coding post-implementation requirement arrivals along three dimensions and, where repository state can be replayed, linking each arrival to a proxy: deletion or replacement of prior agent-authored lines. A requirement's arrival is followed by roughly twice as much invalidation as matched non-requirement edits, robust to user-turn and net-deletion checks, though not demonstrated as causal. This burden shows no detectable decline over a session and no detected association with operation type once multiplicity is accounted for; several intervals remain wide. A controlled experiment shows delayed disclosure relocates implementation post-reveal, while advance warning produces no detected effect on overwriting. These results establish late requirement emergence as a measurable source of code invalidation.

2026-09-02T18:02:18Z Bowen Jiang Haowei Cheng Yuhong Fu Anne Koziolek Jialong Li Weixing Zhang http://arxiv.org/abs/2609.02849v1 2026-09-02T17:33:37Z

Competitive programming has become a key test of large language model reasoning, with international competitions such as IOI and ICPC representing its most challenging settings. We present an end-to-end specialization pipeline combining large-scale problem curation, synthetic reasoning traces, supervised fine-tuning (SFT), and reinforcement learning (RL). Using 22,000 curated problems, we train Nemotron-3-Nano-CC (30B-A3B) with SFT and RL and Nemotron-3-Ultra-CC (550B-A55B) with SFT alone. We further introduce GenCorrect, a feedback-driven test-time compute strategy that iteratively generates, evaluates, and refines diverse solutions. On IOI 2025, Nano-CC improves from 130 points to 291 after post-training and to 468 with GenCorrect, exceeding the gold threshold of 438.3 while Ultra-CC reaches 502. Guided by these results, we develop a competition-specific Ultra-CC system and evaluate it prospectively during IOI 2026. Under the same time, internet-access, and submission constraints as human contestants, it scores 535.4 out of 600, exceeding both the gold threshold of 361.12 and the top human score of 498.27. To our knowledge, this is the first AI system to outscore the highest-scoring human contestant on an IOI problem set.

2026-09-02T17:33:37Z Aleksander Ficek Sean Narenthiran Mehrzad Samadi Somshubra Majumdar Boris Ginsburg http://arxiv.org/abs/2609.02789v1 2026-09-02T16:22:11Z

Large language models are paving the way towards innovation by understanding, analyzing, summarizing and generating content in the modern world. Currently there are thousands of LLM projects developed by engineers in open-source repositories. However, whether these LLM projects have underlying patterns or not remains a question. Exploring these underlying patterns will give new dimensions to the developers who aim to develop these LLM projects. In this paper, we propose ShikumiMiner, a static-analysis framework that combines Abstract Syntax Tree (AST) and Control Flow Graph (CFG) features to detect and compare recurring implementation patterns in C++ local LLM codebases. We analyze ten GitHub open-source repositories and classify functions into seven study-specific categories using a multi-label Random Forest model. Studying these patterns can provide useful insights for developers aiming to design LLM applications.

2026-09-02T16:22:11Z Afsana Tasnim Sheikh Motahar Naim http://arxiv.org/abs/2609.02782v1 2026-09-02T16:15:09Z

Context: In Python, type hints allow developers to annotate variables and functions with explicit type information, improving code clarity and reliability. Although type hints are widely available, little is known about how they are adopted and maintained in libraries and frameworks. Objective: We investigate the adoption, usage, maintenance, and rationale of type hints in Python libraries and frameworks. Method: We analyzed 1,000 popular GitHub repositories, identifying libraries and frameworks and extracting their type annotations. We examined annotation coverage, the locations and origins of annotations, their evolution across git histories, and the relationship between developer annotations and types inferred by Pyright. Results: Of the analyzed repositories, 91% of libraries use type hints at least once, although adoption is inconsistent. Among libraries with systematic usage, maintainers prioritize function parameters and return types, with median coverage of 45.8% and 35.9%, respectively, and mainly use built-in types (73.0%). When modified, annotations tend to migrate to more expressive types. Developers annotate members even when Pyright can infer their types, and these annotations often simplify the inferred type. Conclusion: Type hints in Python libraries and frameworks primarily serve as API contracts rather than comprehensive descriptions of implementation details. The findings suggest opportunities for tooling that prioritizes public interfaces, identifies meaningful annotation changes, and supports maintainers in evolving type information.

2026-09-02T16:15:09Z Thiago Roberto Magalhães Fabio Petrillo João Eduardo Montandon http://arxiv.org/abs/2609.02753v1 2026-09-02T15:55:07Z

Python programs pay for their imports at every process start, a cost that is invisible in steady-state benchmarks but dominant for command-line tools, test workers, and serverless cold starts. Python 3.15 adds explicit lazy imports (PEP 810) largely on anecdotal evidence; no systematic measurement of the ecosystem's import cost exists. We present one: the 500 most-downloaded PyPI packages, sampled quarterly over five years of releases, measured under six CPython versions (3.9-3.14) on two platforms (Apple M5/macOS and Intel Xeon/Linux), for 63,431 measurements in total, plus direct measurement of 3.15's global lazy-import mode. Import cost is heavily skewed: half of packages import in under 6 ms, but the 99th percentile is 354 ms, the first import after installation costs 3-22x more (bytecode compilation), and importing a package's submodules costs up to 294x more than the top-level import that benchmarks report. The median package's cost grows only +1.6-2.4%/year, but the mean grows +11-13%/year: growth is concentrated in a heavy tail. Newer interpreters import the same code 1.16x slower on macOS, but not on Linux, and a single point release (3.11.5 vs. 3.11.16) swings cost by 1.34x. Global lazy mode makes import statements essentially free, yet breaks 8 of 414 top packages. Harness and dataset are available on request.

2026-09-02T15:55:07Z 11 pages, 4 figures, 1 table Trinath Sai Subhash Reddy Pittala http://arxiv.org/abs/2609.02624v1 2026-09-02T14:03:09Z

Assessing vulnerability detection tools for smart contracts requires datasets with known ground truth, yet such datasets are scarce and difficult to build by hand. We propose an approach that uses Large Language Models (LLMs) to automatically inject vulnerabilities into Solidity smart contracts, and demonstrate it in a case study targeting 49 vulnerability types from OpenSCV. Injected contracts are validated through a multi-step pipeline checking compilation, execution, business logic, and the presence of the intended vulnerability. Applied to real-world contracts from SmartBugs, LLMs generate nearly 1,000 candidate variants; after deduplication and validation, 32 confirmed vulnerable contracts spanning 25 vulnerability types survive (a 16.58% survival rate). Surviving contracts concentrate in structurally simpler targets and vulnerability types with localized syntactic patterns. We report practical challenges including LLMs' non-determinism and the difficulty of preserving contract semantics. We then use the validated contracts to assess three static analyzers, revealing complementary and incomplete coverage profiles. Results show that LLM-based vulnerability injection is feasible, while exposing key limitations in scalability and diversity.

2026-09-02T14:03:09Z Luca Migliaccio Roberto Natella Naghmeh Ivaki Nuno Laranjeiro Marco Vieira http://arxiv.org/abs/2609.02983v1 2026-09-02T13:41:50Z

Pattern-based secret scanners are commonly validated with example-based fixtures that fix one variable: the text surrounding a credential. We introduce boundary-mutation testing to vary that context, generating credentials from each rule's own regular expression, embedding them in realistic source contexts, and classifying outcomes at the rule level rather than the tool level, yielding three detection metrics. Applied to three scanners - a 43-rule open-source scanner, Gitleaks 8.21.2, and TruffleHog 3.82.13 - detection in the primary subject holds at >=0.9976 across ten contexts but collapses to 0.5233 when a credential ends in a hyphen. Five rules are affected: two, with fixed-count quantifiers, fail totally and deterministically; three, with variable-count quantifiers, backtrack and match a truncated credential; an entropy fallback rescues some failures but downgrades their severity. We validate a repair restoring full robustness with no new false positives, and report a caution: a plausible first attempt silently regressed two rules, caught only by re-running the same battery. Gitleaks has an unrelated, source-confirmed defect - a hard-coded terminator allowlist causing total misses for most credential types - while TruffleHog shows no boundary fragility but narrowest coverage. We report marginal, not conditional, failure probabilities: one common token format is structurally immune, another fails once in 64; on 292,527 lines of real code, the false-positive ordering inverts relative to the synthetic corpus. Because per-type detection is deterministic, the comparison unit is ten credential types, not hundreds of samples; no recall difference reaches significance, so we report the null result, not a ranking.

2026-09-02T13:41:50Z 17 pages, 10 figures Shweta Mishra http://arxiv.org/abs/2609.02591v1 2026-09-02T13:34:58Z

Much of agriculture depends on open-source software spanning farm management platforms, cloud services, edge gateways, embedded systems, and field-deployed sensors, forming a domain-specific software supply chain that has drawn little empirical security attention. It is unknown whether this ecosystem's supply chain security posture differs from that of comparable non-agricultural software, and if it does, whether the difference reflects the agricultural domain or the size and maturity of the projects within it. As a step towards securing agricultural open-source software, we present AgOSS, a dataset of 66 repositories across six architectural categories. We assess supply chain security within the dataset via OpenSSF Scorecard, governance metrics, SBOM-based dependency analysis, and KEV matching, and compare against matched non-agricultural controls. We report two findings. First, governance is largely independent of inherited dependency risk. Scorecard tracks community activity, but we detect no association with vulnerability count, density, or known-exploited count, and in regression the exposure signal loads on architectural category rather than domain. Second, agricultural projects score far lower on raw Scorecard, but size and maturity confound the gap: the residual loses significance under matching and regression. Securing this ecosystem means investing in contributor capacity and dependency hygiene, not agriculture-specific controls.

2026-09-02T13:34:58Z Vatsal Dudhaiya Mikhail Golovenchits Aryan Banerjee James C. Davis http://arxiv.org/abs/2609.02272v1 2026-09-02T08:21:03Z

Faithfully translating research papers into repository-level implementations remains challenging because papers often describe methods at a high level, leave implementation assumptions implicit, and require generated repositories to preserve method logic, evaluation protocols, and cross-file consistency. Despite recent advances in paper-to-code agents, their intermediate outputs are often presented as free-form plans or summaries that downstream coding agents may ignore, reinterpret, or compress, leading to algorithmic simplification and inconsistent repository structure. To address these challenges, we introduce PaperCompiler, a paper-to-code generation framework that compiles paper-grounded evidence into explicit repository-level implementation specifications. PaperCompiler grounds implementation-relevant evidence while preserving source provenance and distinguishing paper-supported, inferred, externally delegated, and unresolved information. The resulting specifications encode non-degradation requirements, ownership assignments, cross-file dependencies, and file-level constraints. Repository generation proceeds under these compiled specifications while retaining flexibility over local engineering choices not fixed by the paper. PaperCompiler outperforms strong baselines on Paper2CodeBench, achieving a 13.8% relative improvement in reference-based fidelity (from 3.64 to 4.15) and reducing high-severity evaluator critiques (from 13.2% to 6.1%).

2026-09-02T08:21:03Z 9 pages Yunhao Liu Hong Phuc Pham Jaehong Yoon http://arxiv.org/abs/2609.02248v1 2026-09-02T07:55:20Z

Prompt engineering is increasingly used across Software Engineering (SE) activities, including requirements analysis, coding, testing, documentation, repository analysis, and planning. Yet prompts and related instruction artifacts are often created and evolved through task-specific and informal practices, with limited support for their systematic evaluation, management, traceability, and governance. To examine how SE can contribute to the maturation of these practices, we organized a structured community discussion at the First International Workshop on Empirical Prompt Engineering for Software Engineering (PROMPT-SE), co-located with EASE 2026. Participants discussed current prompting practices, challenges to their adoption and evaluation, and future directions for integrating prompt engineering into software development. We synthesized these discussions into five areas: prompt artifacts and standardization; evaluation and benchmarking; lifecycle integration; human-AI collaboration and skills; and governance, privacy, and technical debt. Based on these areas, we outline a research agenda to move prompt engineering from predominantly ad hoc interactions toward more systematic, maintainable, evaluable, traceable, and governable SE practices.

2026-09-02T07:55:20Z Vincenzo De Martino Giovanna Broccia Fabiano Pecorelli Jennifer Horkoff Riccardo Coppola Antonino Ferraro Quim Motger Emma McKenzie Shahbaz Siddeeq http://arxiv.org/abs/2609.02067v1 2026-09-02T03:49:37Z

Scientific benchmarks are commonly built by domain experts who write tasks and cross-check one another's work, or who adapt existing material from textbooks, published papers, and online resources. These routes can produce strong evaluations, but they require substantial per-item labor. Language models can reduce this repeated work by proposing candidates quickly. The remaining problem is acceptance. We target scientific questions whose answers require computations with specialist software rather than unaided reasoning alone. A candidate is invalid if its script fails or returns a different answer, or trivial if a model answers it without the software. We present ToolGate, which treats every generated item as a proposal and keeps it only if three gates pass. First, an executable solution script must reproduce the proposed answer when run with the scientific software. Second, randomized no-tool screening rejects candidates that models can already solve from the prompt alone. Third, a tool-using agent must solve each survivor within a fixed time limit. We instantiate ToolGate in FEniCSx with 500 generation attempts. The local-verification gate retains 478 candidates. For final reporting, we rescreen this pool after generation: two randomized no-tool screens exclude 222 from the reported pool, and direct GPT-5.5 API calls at medium reasoning (the API default) exclude another 121. Of the remaining 135, a GPT-5.5 Codex CLI agent with access to FEniCSx solves 130; exact deduplication leaves 128 unique protocol survivors. ToolGate turns repeated answer checking and difficulty screening into an auditable process while leaving domain design and final review to experts.

2026-09-02T03:49:37Z 7 pages, 2 figures Ke Zhang Yankang Liu Roya Zandi Maziar Raissi http://arxiv.org/abs/2609.01865v1 2026-09-01T20:55:17Z

Embedding-based code retrieval is a core component of coding agents and retrieval-augmented code generation, where retrieving correct code matters more than retrieving lexically similar code. Existing code-retrieval benchmarks do not plant controlled, execution-verified single-edit variants of each query's canonical implementation in the search pool, leaving the question of whether embeddings can functionally discriminate correct from near-clone-but-incorrect code unanswered in a retrieval setting. Resolving this requires a benchmark whose search pool itself contains the relevant counterfactuals -- execution-verified buggy variants near-identical to each canonical -- so that a retriever's rank ordering can be directly tested for functional discrimination rather than topical or identity overlap. We introduce ExecRetrieval, 939 Python tasks each paired with one execution-verified canonical implementation and up to four execution-verified buggy distractors, each generated by a mechanical mutation making a single targeted edit, and evaluate 23 dense embedding configurations plus BM25 under provider-native invocation with paired McNemar tests and query-level bootstrap intervals. With near-clone counterfactuals in the pool, the top hosted system reaches exec@10 = 1.00 but only exec@1 = 0.331; rank-1 misses are paired buggy variants 91.5-99.4% of the time across the four leading systems, and the canonical scores below at least one of its four paired distractors in 67-78% of queries on the leading systems. The full dataset, execution oracle, embedding matrices, environment snapshot, and pairwise statistical tests are released at the URL in Appendix D.

2026-09-01T20:55:17Z Accepted to EMNLP 2026 (Main Conference). Camera-ready version. 17 pages, 6 figures, 8 tables Aaryan Kapoor Md Abdullah Al Hafiz Khan http://arxiv.org/abs/2609.01834v1 2026-09-01T20:08:04Z

As enterprise platforms transition to conversational reasoning interfaces, the stateless nature of LLM APIs creates an architectural gap. While statelessness enables horizontal scalability for AI providers, it forces client applications to manage the entire burden of conversational state and semantic memory. The work identifies the Hydration Proxy Pattern, an architecture that decouples session persistence from the reasoning engine. The framework ensures platform sovereignty over conversational data while enabling secure, multi-stage semantic grounding. We further propose the Context Stabilization Mandate to resolve the tradeoff between sovereign state management and KV caching.

2026-09-01T20:08:04Z 3 pages, 1 table. Presented at the SAO workshop at the 1st ACM Conference on AI and Agentic AI Systems (ACM CAIS 2026) SAO Workshop at the 1st ACM Conference on AI and Agentic Systems (ACM CAIS 2026) Joseph Axisa http://arxiv.org/abs/2609.01781v1 2026-09-01T18:50:26Z

Persisted machine-learning models can remain byte-identical while the software environments in which they are loaded evolve, creating a verification problem that artifact integrity checks alone cannot expose. This paper presents Modelstamp, a lightweight Python persistence library for verifying artifact integrity and represented runtime-environment state before deserialization. At persistence time, Modelstamp associates a serialized artifact with a sidecar JSON manifest containing a SHA-256 digest, runtime metadata, and installed versions from a bounded tracked-package set; a separately recorded model-relevant subset determines which package versions participate in drift comparison. Optional HMAC authentication supports workflows in which the producer and verifier share a secret key. At verification time, the artifact and represented current environment are checked against this recorded evidence before the model is deserialized. Modelstamp is evaluated using 14 controlled environment-drift scenarios, eight controlled trust-boundary scenarios, and an artifact-size scaling benchmark from 10 MiB to 1 GiB. The controlled drift experiments behaved as specified across relevant dependency changes, unchanged environments, and unrelated environmental changes, including broader noise controls. The trust-boundary experiments similarly confirmed both intended detections and expected limitations, including shared-key forgery and replay. Median verification time increased from 0.032 s at 10 MiB to 3.334 s at 1 GiB, with measured throughput of approximately 307-312 MiB/s in the benchmark environment. These results characterize Modelstamp as a complementary pre-deserialization reference-state verification control rather than as a replacement for dependency-management systems, malicious-model detection, safe deserialization, or public publisher authentication.

2026-09-01T18:50:26Z 18 Pages, 2 Figures, 7 Tables, Modelstamp v0.1.4 Source code: https://github.com/AnaghaDhekne/modelstamp Anagha Dhekne http://arxiv.org/abs/2609.01769v1 2026-09-01T18:39:30Z

Automated program repair (APR) research has been constrained to design time. Current techniques localize and fix bugs in RTL or HLS designs before a chip reaches production. Once a hardware vulnerability surfaces post-silicon, the patch must be manually generated: existing automation addresses patch deployment but not patch synthesis. We study the feasibility of extending a dictionary-guided, localize-synthesize-validate APR methodology originally developed for RTL repair to this firmware layer. An automated commit-clustering miner surfaces recurring fix templates across the EDK II (UEFI) firmware repository's full commit history without depending on known CVE identifiers, recovering all three known CVE-fix campaigns and surfacing two additional candidate bug families. Grounded in real fix evidence, we build four independent localizers: missing speculation barriers in C (CVE-2017-5753, Spectre v1), missing bounds checks before array writes in C (a decompression library CVE), missing Return Stack Buffer stuffing in x86 assembly (CVE-2017-5715), and missing integer-overflow guards in Hand-Off Block creation code (surfaced by the miner itself). All four achieve 100% recall; precision ranges from 2.1-15.5% on the C families to 100% on the assembly and HOB families. Root-cause analysis of the C-family false positives attributes 77-90% to two intra-procedural causes, isolating the inter-procedural alias-analysis gap as a measured 15-20% rather than an estimate. A held-out test confirms Spectre v1 localization holds at 100% recall on unseen files; a fifth, independently built dictionary entry (CVE-2018-3630) shows the methodology extends to a new bug signature at low cost; and a naive syntactic baseline recalls at most 14% where our detector recalls 100%. We frame these results within a broader research agenda for a unified hardware-to-firmware correctness lifecycle.

2026-09-01T18:39:30Z Maisha Mastora Dean Sullivan http://arxiv.org/abs/2609.01758v1 2026-09-01T18:25:01Z

Large Language Models (LLMs) are increasingly used for software vulnerability detection, but their performance depends on how source code is represented in the input. Most prompting approaches use source code in its original form, while some works propose the use of structured representations. Abstract Syntax Trees (ASTs) are one of the most popular approaches, but AST verbosity increases input size relative to source code, making them hard to fit within some LLMs context windows. This paper investigates Behavior Trees (BTs) as an alternative intermediate representation for LLM-based vulnerability detection. BTs encode control flow, conditions, and executable actions more compactly than ASTs, making them a natural candidate when token count is a constraint. First, we propose a preprocessing stage that parses Java source code into ASTs and then converts them into BT representations. We then compare vulnerability detection performance across 460 Java samples from the Juliet Java test suite, using three input representations: raw source code, AST, and BT. All experiments use a single quantized local LLM, Mistral Small 3.2 24B (Q4_K_M). Our results show that using BT representations improves recall on short code samples, while raw source code achieves higher precision. On longer samples, BTs improve overall performance over the original representation and fit within the context window, whereas many ASTs exceed the context limit. These findings suggest that BTs can provide a compact and useful structured representation for vulnerability detection with quantized, locally deployable LLMs.

2026-09-01T18:25:01Z Enna Basic Alberto Giaretta http://arxiv.org/abs/2609.01736v1 2026-09-01T18:07:09Z

Large language models (LLMs) augmented with external tools have demonstrated remarkable capability in solving complex real-world tasks. However, existing approaches suffer from two key challenges: brittle multi-step and multi-turn reasoning caused by incompatible tool output types and API schemas, and performance degradation under large tool catalogues. To address these, we introduce \textbf{Tool Primitives}, a design that replaces rigid API schema-based invocation with natural language as the interface for tool calling, where each tool is wrapped with an LLM interface that handles schema resolution and execution internally, enabling natural inter-tool communication for nested and multi-turn tool calling. Building on Tool Primitives, we host \textbf{ToolFace}, a centralized repository of 25,519 functions from which LLMs dynamically retrieve only the relevant tools at inference time, eliminating the need to enumerate raw API schemas in context. To orchestrate Tool Primitives and ToolFace reliably in complex settings, we further propose \textbf{HEART}, a \textbf{H}arness \textbf{E}ngineering framework via \textbf{A}gent-native, \textbf{R}eusable \textbf{T}ool Primitives, comprising a Planner, Router, and Verifier that jointly support dynamic tool invocation planning, multi-step execution, and feedback-driven recovery. Experiments on five benchmarks demonstrate that HEART outperforms SFT-based models by $10\%$ on average and surpasses GPT-5.4, Claude-4.6-Sonnet, and Gemini-3.1-Pro by $6\%$ on average while reducing API cost by up to $85\%$. On 50 real-world tasks, HEART achieves $84\%$ task completion, $3.8\times$ the average of three frontier commercial models ($22\%$).

2026-09-01T18:07:09Z 21 pages Haibo Jin Suijin Wang Xucheng Yu Haojing Luo Haohan Wang http://arxiv.org/abs/2609.01603v1 2026-09-01T17:59:46Z

Evaluating software engineering agents on realistic benchmarks is costly, since each task may require multi-step code exploration, modification, and test execution. Existing efficient evaluation methods select representative subsets to estimate full-benchmark performance, but are largely result-only: they fit historical pass/fail response matrices or static task semantics, discarding how agents solve problems. We propose PTA-IRT, a Privileged Trajectory-Aware Item Response Theory framework that fuses process and outcome signals. Historical execution trajectories supply process-level evidence beyond pass/fail, such as explored context, attempted edits, and solving paths, which PTA-IRT uses as privileged information for calibration subset selection and ability estimation. Under low calibration budgets, PTA-IRT consistently outperforms prior IRT baselines on score and ranking recovery across four SWE benchmarks. Code and data are publicly available at https://github.com/DeepSoftwareAnalytics/PTA-IRT.

2026-09-01T17:59:46Z Under review Kefeng Duan Dewu Zheng Yanlin Wang Xiwen Wang Ensheng Shi Xilin Liu Yuchi Ma Jiachi Chen Mingwei Liu Zibin Zheng http://arxiv.org/abs/2609.01601v1 2026-09-01T17:59:39Z

The repository-level code generation task requires synthesizing code that satisfies task requirements while remaining consistent with the target repository context. Since real-world repositories often exceed the input length limits of LLMs, existing approaches commonly adopt retrieval-augmented generation (RAG) to provide repository-specific context. Despite improving repository-context retrieval, existing methods typically provide context as task-level support, without explicitly identifying the critical tokens that require fine-grained repository context during generation. During the autoregressive generation process of LLMs, errors often concentrate at a small number of decisive positions: once such tokens are generated incorrectly, subsequent code may follow an incorrect semantic path and eventually lead to functional failure. We refer to these positions as "critical tokens". In this paper, we propose ACToR, an adaptive critical token-aware retrieval framework for repository-level code generation. ACToR identifies critical tokens during generation and triggers targeted retrieval on demand to provide repository context at these decisive positions. In addition, we design a position-aware weighting method for dense retrievers to prioritize context that is more informative for generation. We evaluate ACToR on two representative repository-level benchmarks, RepoExec and CoderEval. Experimental results show that ACToR consistently outperforms state-of-the-art methods, achieving relative improvements of 8.4% on RepoExec and 15.4% on CoderEval. Beyond performance gains, we systematically quantify the impact of critical tokens, revealing their central role in major generation failures and highlighting the necessity of targeted retrieval strategies. We provide the code and data at https://github.com/DeepSoftwareAnalytics/ACToR.

2026-09-01T17:59:39Z Under review Kefeng Duan Dewu Zheng Yanlin Wang Terry Yue Zhuo Mingwei Liu Jianxing Yu Jiachi Chen Ensheng Shi Xilin Liu Yuchi Ma Zibin Zheng http://arxiv.org/abs/2609.01702v1 2026-09-01T17:22:07Z

Agent-assisted software projects are often reported through demonstrations or aggregate benchmarks that conceal how correctness claims were admitted. This experience report studies RosettaBitcoin, a single-developer project that built twelve separately implemented Bitcoin testnet4 consensus validators, through its immutable 17 June 2026 software snapshot (DOI 10.5281/zenodo.20738249). We analyze the snapshot's tracked SQLite evidence database, curated artifact index, conformance fixtures, validation scripts, blocker records, and version history. At the snapshot, all twelve ports had port-owned 45/45 script-corpus proofs and strict 5,000-block baselines. Nine had canonical clean 50,000-block, 100,000-block, and post-100,000 validation lanes. Java had one 19.86-second near-tip maintenance artifact. No port had an empty-state-to-tip proof, and no port satisfied the project's binary full-node gate; Docker and live-node capability gaps remained. The artifact history also records a 3 h 17 min 57 s Zig scaffold-to-50,000 span, but the observed intervals describe non-equivalent tasks and cannot estimate effort, productivity, or causality. A separate diagnostic supplement preserves evidence that a pure-Mojo cryptographic backend validated fresh state to height 100,000 and resumed to 140,234, agreed on a 45-case shadow comparison, rejected six crafted invalid classes, and was killed by three targeted mutations. That evidence is noncanonical, noncomparable, and class-bounded. The case suggests that explicit failure records, fixtures, port-owned proofs, and validating imports can make agent-assisted systems more auditable. Controlled ablations and external replications are needed to test whether such infrastructure causally improves development outcomes.

2026-09-01T17:22:07Z 8 pages, 1 figure, 3 tables. Software snapshot: 10.5281/zenodo.20738249. Diagnostic supplement: 10.5281/zenodo.22114337 Donavon Guyot http://arxiv.org/abs/2609.01503v1 2026-09-01T16:33:30Z

Learning- and LLM-based software vulnerability analysis is only as trustworthy as the data it is trained and evaluated on, yet that data is rarely examined as a first-class object. We investigate the data behind vulnerability analysis through a dataset-centric taxonomy that separates what an artifact is (code, metadata, patches, tests/PoCs, reasoning, traces), how good it is (realism, label evidence, scale, diversity, leakage, availability), and what it is used for. From a systematically assembled corpus of 1522 papers covering 2016-2026 plus foundational earlier work we deep-code a tiered set of 111 anchor papers, backing every affirmative rubric-graded value with a verbatim span, and we report, per attribute, both how much it has been studied and how well datasets achieve it. The results trace an evidence ladder: executable artifacts are the only major type where 15 of the 24 datasets are both graded real-world and carry labels that received an independent check, while code-sample datasets-the largest category in both the auto-tagged corpus and the anchor set-are the least realistic: 20 of the 41 draw their vulnerabilities from authentic projects or CVEs, but only 3 keep the sample at the unit the code is deployed in, and only 2 do both-though these are coarse component tests, and just one code-sample dataset meets the codebook's stricter full-context real-world grade. Among these, leakage goes unaddressed by 49 of the 90 datasets where it applies, more than a quarter say nothing about availability, reasoning data has arrived only recently and is mostly model-generated, and primary trace corpora remain limited to three datasets, the total after a corpus-wide screen and a full-text check of every candidate it surfaced, with further datasets releasing traces secondarily behind benchmarks and harnesses.

2026-09-01T16:33:30Z Yu Nong Yao Du Tianxiang Xu Haipeng Cai http://arxiv.org/abs/2609.01437v1 2026-09-01T15:45:33Z

As agents move from research prototypes to deployed tools, their capability increasingly depends on model-external execution infrastructure, commonly termed the agent harness. Changing this harness while holding model weights fixed can substantially alter task performance. Current agent evaluations typically report downstream performance under a chosen harness, leaving a model's ability to develop the harness itself comparatively underexplored. We introduce HarnessDev, a benchmark that shifts the unit of evaluation from task outputs to runnable infrastructure. HarnessDev covers two stages. In Creation, the agent starts from a minimal seed and a small number of cases, then builds a complete execution system. In Evolution, it starts from its own created harness and iteratively revises it using downstream execution feedback, with the goal of improving benchmark performance. We then evaluate each constructed harness on capability (task success on held-out benchmarks) and efficiency (execution-token cost). The reported Creation results cover six creator LLMs, four domains, and five downstream benchmarks totaling 2,207 unique downstream instances, with hidden evaluation tasks withheld from development. We find that generated harnesses remain substantially behind mature human-engineered references on code and on search and research, while matching or exceeding the selected references on writing and machine-learning experimentation, with large variation in execution cost. Evolution produces some performance gains, but they are unstable and transfer only partially to held-out tasks. Experiments with a fixed runtime model further show that the gains depend strongly on the model executing the harness, indicating limited transfer across models.

2026-09-01T15:45:33Z Project page: https://self-developing-agents.github.io/ Yuhao Wu Jingyuan Zhang Jiajun Shi Xinping Lei Qingshui Gu Yuxuan Zhang Zexuan Wang Chen He Chen Huang Maojia Song Zhiyuan Zeng Shaowen Wang Jinkai Liu Yunfeng Shi Jiaheng Liu Shen Yan Wenhao Huang Ge Zhang Wenxuan Zhang http://arxiv.org/abs/2609.01271v1 2026-09-01T14:04:15Z

Agentic software engineering benchmarks are typically summarized by nominal category labels such as "bug fix" or "feature implementation," yet benchmarks carrying the same label are built through very different curation pipelines. A label thus reveals little about the engineering work a benchmark demands. We introduce the Spread--Novelty--Centrality (SNC) profile, a three-axis characterization of the demands of repository-level coding tasks, grounded in empirical software engineering research. We apply the profile to five widely used benchmarks and 14,922 trajectories of two model families at three scales, and report three findings. (1) A label is an unreliable proxy for task demands, as every pair of benchmarks is statistically separated on at least two SNC axes, and the separations trace back to specific curation decisions. (2) Agent behaviour reveals demands that the human-written gold solution cannot. Agents produce larger solutions than the gold where problem statements withhold hints and smaller ones where curation inflates the gold. How a task is phrased shapes what an agent produces. (3) Task demands correlate with success uniformly, with resolved runs concentrating in the low-SNC region for every family and scale, whereas the behavioural signatures of success are family-specific. Claude succeeds by matching the scope of the gold solution, and its parity share on files rises from $0.17$ at the smallest scale to $0.54$ at the largest. Qwen succeeds by exceeding the gold scope at every scale, and editing too little marks failure for both families.

2026-09-01T14:04:15Z Radin Shayanfar Keheliya Gallaba Ahmed E. Hassan http://arxiv.org/abs/2609.01236v1 2026-09-01T13:35:29Z

Large language models have shown promising capabilities in code refactoring, but existing approaches remain limited to method-level tasks. In this paper, we envision LLM-based refactoring as a continuous component of software maintenance rather than a tool invoked only for occasional manual refactoring. Under this vision, AI agents continuously monitor, evaluate, and improve codebases against explicit and evolving notions of software quality. We present a roadmap organized around five dimensions: the multi-objective optimization problem, quality definition and evaluation, multi-timescale integration of heterogeneous signals, architecture and design pattern, and trust in autonomous refactoring. We further identify integration into continuous delivery pipelines and cost considerations as cross-cutting concerns. For each dimension, we analyze the underlying challenges and pose open research questions. These dimensions define a research agenda for advancing autonomous refactoring from isolated code improvements to system-level quality maintenance.

2026-09-01T13:35:29Z The paper was accepted in ICSME 2026 in the Visions and Emerging Results Track Xin Sun Daniel Ståhl Kristian Sandahl Christoph Kessler http://arxiv.org/abs/2609.01187v1 2026-09-01T13:01:45Z

A single vulnerability in a widely used library can cascade through millions of dependent applications, yet more than half of vulnerability database entries contain missing or incorrect affected-library information. Existing automated approaches neglect the relational structure of vulnerability databases, treating identification as an isolated text retrieval problem. In this paper, we propose Athena, the first graph-based approach for vulnerability affected library identification. Athena models vulnerability databases as a knowledge graph and reformulates the identification problem as knowledge graph completion (KGC). It comprises three key modules: a Modeling module that constructs a security knowledge graph integrating CVEs, libraries, CWE weakness types, CPE products, and software ecosystems; a Completion module that applies a modular KGC backbone to predict missing affected libraries for a given CVE via link prediction; and a Re-ranking module that retrieves KGC candidates and rescores them using a fine-tuned LLM augmented with knowledge graph embeddings, jointly leveraging structural and textual information. Our experiments on VulLib demonstrate that Athena significantly outperforms four state-of-the-art baselines, achieving a 32% improvement in Avg. F1 over the best baseline (i.e., VulLibGen). Notably, our KGC backbone with only 110M parameters already surpasses VulLibGen's best configuration at 7B parameters, demonstrating the effectiveness of graph-based modeling; the re-ranking module then provides substantial further gains, consistently outperforming the best baseline across all evaluated LLM backbones.

2026-09-01T13:01:45Z Accepted to EMNLP 2026 (Main Conference) Phong Trinh Duy Trang Dang Yen Hung Nguyen-Huu Bach Le Quyet-Thang Huynh Dieu Hoang Vu David Lo Thanh Le-Cong http://arxiv.org/abs/2609.01186v1 2026-09-01T13:01:39Z

The Common Vulnerabilities and Exposures (CVE) database catalogs vulnerability claims in hard- and software, among them those pertaining to blockchain programs a.k.a. smart contracts. We present CVE-Smart-Contracts, a curated dataset of CVE records up to July 2026 referring to Ethereum smart contracts. The dataset contains the vulnerable artifacts (source code and runtime bytecode), labels according to three taxonomies, and function-level locations. The retrieval of CVE records, collection of additional evidence, validation of the correspondence between records and artifacts, label assignment, and vulnerability localization are automated, leaving 15% to manual analysis. The dataset does not validate the original vulnerability claims, but marks a few records obviously wrong as `refuted'. For the sake of reproducibility, all external inputs are retained, so that rerunning the pipelines results in the same outputs. The dataset comprises 491 records linked to deployed contracts, 26 referring to projects (mostly libraries), 45 without validated artifacts, and six records with refuted claims. The dataset supports empirical security research, in particular the evaluation of code analysis and repair techniques.

2026-09-01T13:01:39Z Monika di Angelo Gernot Salzer http://arxiv.org/abs/2609.01106v1 2026-09-01T11:47:39Z

When a hint turns a failing generated program into a passing one, does it provide missing information or merely steer the model toward a solution it could already produce? We test these hypotheses on HumanEval+ and MBPP+ using executable evaluation. For Qwen2.5-3B-Instruct, adaptive relevant hints rescue 36 of 79 selected failures; an unrelated hint rescues 19, while eight unhinted samples solve 46 and recover 31 of the 36 relevant-hint rescues. Phi-3.5-mini shows the same pattern: relevant hints rescue 42 of 101 failures, an unrelated hint rescues 17, and unhinted sampling solves 57, including 36 of the 42 relevant-hint rescues. Because the hint conditions use different attempt budgets, these comparisons do not isolate a purely semantic effect. Mechanistic tests on Qwen identify a stable activation direction shared by relevant and unrelated hints. Persistently adding this direction yields 14 rescues and 18 regressions, with no detectable net accuracy gain; learned low-rank interventions have a positive but imprecise estimated effect. Full textual specifications solve 22 of 24 context-defined problems, versus 5-11 for tested virtual-KV prefixes. Post-generation hidden-state probes transfer across benchmarks, with pooled AUROC 0.806 and 0.780, but their top-one selection advantage over token confidence is statistically unresolved. Overall, relevant hints can rescue failures, but most rescued solutions are already reachable through ordinary sampling, and the internal interventions tested here do not establish task-general capability transfer.

2026-09-01T11:47:39Z Will Badr http://arxiv.org/abs/2609.01095v1 2026-09-01T11:35:02Z

Extracting physics information from modern particle-physics experiments requires multistage analyses implemented on top of large and highly interconnected software ecosystems. General-purpose large language models (LLMs) often produce unreliable programs for such tasks because a user request alone rarely specifies the required APIs, dependencies, and usage conventions. We organize these software relations before generation and retrieve task-relevant knowledge at inference time. Using the open-source ROOT framework as a representative and reproducible testbed, we evaluate a complete grounding system that combines hybrid retrieval over a heterogeneous software knowledge graph, skill-selected workflow examples, and execution-guided repair. On a benchmark of 275 ROOT tasks, grounding improves first-attempt execution from 58.5% to 76.0% under Claude Code orchestration and from 51.3% to 64.0% under standalone orchestration. Final success increases from 90.5% to 96.0% and from 78.9% to 90.9%, respectively, while the average generation cost per successful task increases by only 1.3% and 3.2%. The gains persist under a strong coding agent, indicating that explicit software knowledge remains valuable even when agentic scaffolding is already in place. Because the method captures software relations common to large codebases rather than facts specific to ROOT or a particular model, it should transfer to other experiment frameworks and proprietary software, especially where documentation is sparse or internal dependencies are complex.

2026-09-01T11:35:02Z 24 pages, 5 figures, 5 tables Yue Sun Tong Liu Yipu Liao Jingde Chen Ke Li http://arxiv.org/abs/2609.01087v1 2026-09-01T11:25:36Z

Context: Accurate alignment between pull requests (PRs) and corresponding issues is crucial for efficient software development and maintaining code quality, as misalignments can reduce traceability, hinder defect localization, and decrease maintainability. Objective: This study aims to improve automated PR-issue alignment classification by leveraging fine-tuned large language models (LLMs) across multiple alignment categories, and conducts interpretability analysis to investigate the effects of PR-issue fields on the predictions of fine-tuned LLMs. Method: Our methodology consists of dataset preparation, LLM fine-tuning, and interpretability analysis. We first extended an existing dataset and applied data augmentation to address class imbalance. GPT-4o was then fine-tuned via instruction tuning, and open-source LLMs including CodeLlama-7B, CodeQwen1.5-7B, StableCode-3B, CodeGemma-7B, and Deepseek-Coder-6.7B were fine-tuned using classification-specific heads. Interpretability analysis using Shapley Additive Explanations (SHAP) was conducted to examine the influence of PR-issue fields on predictions for the best-performing open-source LLM. Results: Fine-tuned LLMs outperformed baseline models, achieving average improvements of 6.15% in accuracy and F1-micro, 14.69% in F1-macro, and 6.15% in recall. CodeLlama-7B emerged as the best-performing fine-tuned LLM overall, while interpretability analysis revealed that code diffs together with issue body and PR body contents exert the greatest influence on predictions. Conclusions: Fine-tuning substantially enhances PR-issue alignment classification, improving both accuracy and efficiency. Interpretability analysis provides actionable insights into the dataset features driving alignment decisions, deepening understanding of how LLMs reason over software artifacts.

2026-09-01T11:25:36Z 31 pages, 12 figures. Submitted to Springer Empirical Software Engineering (EMSE), Special Issue on Software Analysis, Evolution, and Reengineering (SANER 2025) Mustafa Yasir Altunhan Hüseyin Özgür Kamalı Eray Tüzün http://arxiv.org/abs/2609.00854v1 2026-09-01T07:48:51Z

Fault localization can focus a code model's repair on the statements a failing test implicates, but a targeted edit may succeed merely because it is small, and a second model call may succeed without using the failure at all. We separate these explanations with three arms applied to the same failed candidate: blind whole-solution resampling, spectrum-based localization followed by suspect-span infilling, and same-length infilling at a disjoint random code span. Across three frozen 26-32B models, three benchmarks and 488 failing candidates, plus a separately declared 24B fourth model from a third family, three results follow. First, localization is rarely available: only 9.0% of failing candidates expose a failing public test with a usable spectrum. Second, among the 177 candidates localizable from a strong suite, localized infilling loses decisively to blind resampling at a matched attempt count (3:40, p = 3.0 x 10^-9), opposite to our hypothesis; the loss replicates in a third family at -11.3 points (95% CI [-16.6, -6.8]), and widening the edit does not rescue it. Third, against the random-span placebo localized infilling leads pooled (11:1, Holm-adjusted p = .019), but that lead resolves in no individual model under the analysis our shipped plan designates primary (best Holm p = .087), so we report the location effect as suggestive rather than established. Re-pricing attempts as tokens narrows but does not overturn this: a span attempt spends 21.7 generated tokens against 371.1, yet 16 localized attempts reach 6.8% while one blind attempt already reaches 10.1%. Infilling reproduces the removed span verbatim in 48.9% of attempts, which is why more budget does not help. We restrict every localization conclusion to the 24-32B models tested.

2026-09-01T07:48:51Z Anik Jha http://arxiv.org/abs/2609.00838v1 2026-09-01T07:37:55Z

Test-set accuracy is silent on two issues that matter when deploying autoregressive neural sequence models: how much probability mass the system under test (SUT) places on constraint-violating alternatives that are reachable under sampling and what fraction of the input population satisfies a domain requirement. We answer both with probabilistic model checking. The pipeline extracts a discrete-time Markov chain (DTMC) from the SUT's token-by-token generation, verifies formal PCTL specifications with the PRISM model checker, and aggregates the per-input verdicts into a coverage curve over the input space. A soundness theorem establishes the DTMC as an under-approximation, so every verdict yields a certified interval on the SUT's true reachability probability. The coverage built from those verdicts is, therefore, conservative by construction. A counterexample-guided abstraction refinement (CEGAR) loop adaptively tightens the interval, and a maximum-likelihood algorithm extracts the most probable falsifying trace. Two case studies exercise the pipeline. On a GPT-2 computer-aided process-planning (CAPP) model with 100% test accuracy, the pipeline quantifies the probability mass greedy decoding hides, but that is reachable with sampling; and identifies the smallest training fraction at which an ordering requirement holds population-wide, neither of which test accuracy can report. We then verify the SMILES molecular generator with a 50x larger vocabulary. The only change is an external chemical-validity oracle, and the pipeline identifies the gap between structural completeness and chemical validity.

2026-09-01T07:37:55Z 38th International Conference on Testing Software and Systems (IFIP ICTSS 2026) Helge Spieker Dennis Gross Arnaud Gotlieb http://arxiv.org/abs/2609.00834v1 2026-09-01T07:35:00Z

Modern Text-to-SQL systems often follow generate-execute-select pipelines, generating multiple candidate queries then selecting the best one. Listwise selection, by jointly comparing multiple candidates, has been widely adopted, but fine-tuning listwise selectors is costly. We thus propose a fine-tuning-free listwise selector. We replace two major fine-tuning objectives with inference-time strategies: (1) learning selection criteria as ordering and (2) mitigating positional bias. First, we build reusable structured memories instead of learning selection behavior as model parameters. Given a question, MaP-SQL retrieves memories distilled from training data that encode how natural language maps to schema elements, SQL operations, and expected outputs. These memories serve as explicit decision criteria for evaluating candidates in a listwise manner. Second, to mitigate ordering bias of listwise selectors, we aggregate rankings across multiple input permutations, with inference cost optimized by execution results and pointwise scoring. Our approach improves selection accuracy while maintaining efficiency and compatibility with existing large language models. Across Text-to-SQL benchmarks, it produces more stable selection without fine-tuning and fewer unnecessary comparisons than existing methods. On BIRD-dev, it outperforms the previous state-of-the-art selector-based method R^3-SQL by 2.02 execution accuracy points on average using the same candidate sets, with 2.92x fewer tokens.

2026-09-01T07:35:00Z Accepted by Findings of EMNLP 2026 Yeonseok Jeong Soyoung Yoon Seongjun Lee Seung-won Hwang http://arxiv.org/abs/2609.00805v1 2026-09-01T07:02:06Z

LLM-based software systems increasingly require effective "evals" as quality gates in the development lifecycle. However, existing work typically addresses individual aspects of eval reliability rather than the full set of practical requirements. We present an end-to-end eval pipeline that combines eval checklist creation, with learned aggregation for checklist responses, to improve agreement across LLM judges and accuracy against human judgments. The framework additionally pro- vides self-consistency, explanations, and prediction uncertainty, and we empirically demonstrate its effectiveness.

2026-09-01T07:02:06Z Emma Thuong Nguyen Abhishek Ghose http://arxiv.org/abs/2609.01669v1 2026-09-01T04:04:25Z

Developers face a challenging problem with no clear solution. Modern software breaches can wreak havoc on businesses and individuals alike. With code vulnerabilities being a leading cause, securing applications must be a priority for developers. Static Application Security Testing (SAST) has the potential to harden applications by assisting in the identification and resolution of security vulnerabilities. Despite this, many development teams have not adopted SAST tools into their environment. In this paper, we survey the recent literature to uncover why some developers are apprehensive towards SAST and identify what specific problems they encounter when using it. We found a variety of usability problems developers face when using SAST. Some are inherent of the tool and ultimately require some level of developer investment while others are tool shortcomings that SAST tool creators must address. Ultimately, we argue that in order to drive widespread adoption and consistent SAST usage, developers will need to embrace that some investment is required. Simultaneously, developers will be more likely to integrate SAST tools into their workflows if the creators of SAST tools simplify many aspects related to tool usage. Surmounting the primary obstacles preventing the adoption of SAST requires full consideration of both the technical and human factors.

2026-09-01T04:04:25Z 6 pages, 4 figures, published at 2024 39th IEEE/ACM International Conference on Automated Software Engineering Workshops (ASEW) 39th IEEE/ACM International Conference on Automated Software Engineering Workshops (ASEW '24), Sacramento, CA, USA, 2024, pp. 161-166 Zachary Wadhams Clemente Izurieta Ann Marie Reinhold 10.1145/3691621.3694947 http://arxiv.org/abs/2609.00579v1 2026-09-01T02:18:20Z

Large language models (LLMs) have shown proficiency in various software engineering tasks, such as code generation and translation. However, a key limitation in their performance may be their (lack of) understanding of programming-language semantics. Even when explicit semantics are given, it remains unclear whether LLMs apply those rules or lean on priors learned during pre-training instead. We study if LLMs lean on priors or given semantics with a novel task--Program Executability Prediction (PrEx)--that asks models to predict whether a program is semantically valid or invalid (and, if invalid, which formal rule it violates) given the program's syntax and operational semantics. Because PrEx requires both valid and invalid programs, we build a dataset with systematically generated invalid transformations derived from valid programs. We evaluate open-source coding LLMs under two semantic formalisms and two semantic shifts across Human-Written, LLM-Translated, and Fuzzer-Generated program splits. Our findings show that LLMs lean on pre-training priors rather than systematically applying the given rules, performing especially poorly on modified semantics and degrading further as program complexity increases. PrEx is available at https://github.com/EngineeringSoftware/prex.

2026-09-01T02:18:20Z Accepted at LMPL 2026 Lara Marinov Aditya Thimmaiah Jayanth Srinivasa Junyi Jessy Li Milos Gligoric 10.1145/3843750.3843842 http://arxiv.org/abs/2609.00568v1 2026-09-01T02:04:16Z

Code generation aims to automatically generate source code from task requirements and has attracted significant attention with the rapid advancement of large language models (LLMs). Despite remarkable progress, LLMs often struggle to generate correct code for complex software engineering tasks because task descriptions are frequently incomplete, ambiguous, or lack critical contextual information. Existing approaches primarily improve the capabilities of coding agents through more sophisticated tools, skills, and workflows, while largely overlooking the quality of the task requirements themselves. To address this limitation, we draw inspiration from software requirements engineering and propose WiseSpec, a novel requirements-driven agent framework for repository-level code generation. WiseSpec automatically constructs structured and information-rich requirements, assesses their quality through execution-based evaluation, and iteratively refines them to better guide code generation. Experimental results show that WiseSpec consistently outperforms all baselines, achieving an average improvement of 13.17% in %Resolved.

2026-09-01T02:04:16Z Accepted by ASE 2026 (SRC) Zhao Tian http://arxiv.org/abs/2609.00546v1 2026-09-01T01:32:22Z

Agent systems are commonly described by the model and harness that currently produce their behavior. That boundary is useful for one execution but underspecifies a long-lived agent that may change models, orchestration harnesses, interaction sessions, and host servers while retaining one identity, memory, and executable code lineage. We present a runtime-independent architecture for persistent agents. A continuity-bearing substrate $P_t=(I_t,M_t,B_t)$ contains an architectural identity representation, private durable memory, and a versioned software body. A replaceable deployment binding comprises an execution substrate $E_t=(R_t,H_t,D_t)$, which supplies a reasoner, harness, and host, and a set of interaction surfaces $S_t$, such as chat, API, or user interface bindings. A deployed execution is $A_t=P_t\triangleright(E_t,S_t)$; changing either replaceable layer is migration, not agent creation, when an authorized protocol preserves attributable lineage and transfers continuation authority within a governed deployment boundary. We define six continuity invariants and a quiesce--checkpoint--validate--bind--rehydrate--resume protocol. Enoch realizes the design as a reusable body plus private installed identity, memory, workflow state, and continuation authority, with infrastructure dependencies behind versioned provider contracts. A clean-room run of the frozen public commit passes 833 core tests and 92 provider and library tests executed separately from the core suite; deployments have exercised reasoner-version, interaction-surface, and host-machine substitutions while retaining continuity-bearing state. This evidence supports mechanical substitutability and authorized system continuity, not behavioral invariance or exhaustive pairwise evaluation. The downstream measurement question is whether an authorized continuation still recalls, composes, and enacts its identity.

2026-09-01T01:32:22Z 8 pages, 2 figures, 3 tables. Reference implementation and artifacts: https://github.com/our-ark/enoch Zhenyu Zhao Independent Researcher Roy Zhao Paul G. Allen School of Computer Science & Engineering, University of Washington