traceability

EDMA提出了一种基于逻辑熵的多模态问答方法，将答题过程形式化为熵减迭代，并量化各模态对答案的贡献，在多个MMQA基准上显著优于Chain-of-Thought等基线。

The article discusses the challenges of managing autonomy for AI agents in software development, raising questions about permissions, traceability, and the role of human oversight.

EvidenT is a lightweight pipeline that enhances evidence groundedness and traceability in enterprise retrieval-augmented generation (RAG) systems, improving gold-source hit rate by 29% without model retraining.

The article explores the key challenges in transitioning LLM application ideas to production in the EU, covering legal compliance, traceability, abuse prevention, and accountability, and seeks concrete examples from practitioners.

The article poses a question about where accountability should start when AI agents use tools and APIs, debating between tracing every step or focusing on final outcomes.

The paper proposes TraceSQL, a lightweight and traceable verification model for text-to-SQL systems that uses explicit diagnostic features to estimate answerability without reference queries, achieving improved performance over existing baselines on the BIRD benchmark.

This paper presents a triple-robustness analysis of RAG for multi-hop traceability, varying embedder, corpus, and judge across thousands of runs. It finds that over-citation is architecturally universal in GraphRAG but its faithfulness consequences are corpus-conditional, and LLM-as-judge faithfulness is fragile across retrieval states.

Presents HaGen, an AI-assisted approach that generates traceable hazard scenarios for aviation operational safety analysis from NASA's ASRS reports, providing plausible structured hypotheses and narratives with evidence links.

This paper introduces TraceMAS, an interactive demo system for traceable multi-agent forecasting that organizes agent outputs around causal-loop diagrams to make the evidence-to-forecast process inspectable, demonstrated on crude oil price forecasting.

The paper introduces DocTrace, a hierarchical framework for long document visual question answering that casts the task as explicit evidence graph reasoning. It achieves state-of-the-art results on three benchmarks while enabling traceable evidence provenance, outperforming Qwen3-VL-8B-Instruct by 11-14 points.

The author introduces ARGO, an open-source Agent Loop framework for Cursor, GitHub Copilot, and OpenCode that combines an ArchiMate-structured intent graph with GraphRAG retrieval to make AI coding delivery traceable. It separates intent, architecture, and implementation responsibilities, with two-level acceptance tests to avoid treating every failure as a code bug.

This paper introduces evidence-ledger adjudication, a workflow for claim-evidence traceability in AI-assisted writing, evaluated on a blind benchmark from AVeriTeC, CLIMATE-FEVER, and SciFact, showing agent-based methods outperform baselines.

Humanly is a configurable writing platform that records the writing process to provide traceable evidence of human-AI collaboration, with features like sealed certificates and anomaly detection.

This paper presents the first rigorous study of how LLM watermarking schemes affect medical performance, evaluating five watermarks across multiple LLMs and VLMs on clinical reasoning tasks. The authors find that watermarks can cause degradation in medical text quality, including hallucinations and lexical corruption, which are masked by general-domain benchmarks.

ArcKit is an open-source enterprise architecture governance harness that works with AI coding assistants like Claude Code and GitHub Copilot to manage governance workflows including requirements, design reviews, and traceability.

This paper proposes a multi-level validation and traceability framework for AI-generated telescope scheduling decisions, integrating data reference validation, logical consistency checks, and observational constraints to improve executability and reliability in high-reliability astronomical observation tasks.

ModSleuth is a new tool that traces the dependencies of modern LLMs, revealing that models like OLMo 3 and Nemotron 3 rely on hundreds of other models and datasets, highlighting the shift from human-only to AI-generated training data.

The article critiques current AI memory systems as mere write-only logs that lack the ability to be corrected, updated, or traced to their source, arguing that true memory requires a governance layer.

A developer built a fully traceable and forkable research agent using Active Graph and monid_ai, ensuring every claim is natively traced to its source, and got it working in about 30 minutes.

The author argues that current agent observability provides a trace of actions but lacks runtime justification for why actions were permitted, which is critical for production deployments involving money, data, or communications.