retrieval

The paper introduces FABench, a scalable backtesting environment for forecasting agents with an automated pipeline that discovers events, generates questions, and resolves outcomes to avoid training-cutoff leakage. Evaluating agents on eight recent models, it finds retrieval consistently improves Brier scores and lifts leading agents to Polymarket market consensus, while revealing recurring evidence-weighting failures.

Meta's DecisionRank paper introduces item-space on-policy distillation to transfer a generative LLM teacher's ranking knowledge into a single-token decision-model reranker, achieving state-of-the-art results on TREC DL, BEIR, and BRIGHT while running up to 24.5x faster than its generative counterpart and surpassing its 27B teacher at just 1.7B parameters.

A preprint tests five agent memory systems (Graphiti, Zep, mem0, langmem, cognee) for policy revocation, finding no system reliably prevents agents from retrieving and acting on superseded harmful policies, with revoked records sometimes ranked first and unsafe actions taken in over 40% of trials.

LangChain Community Spotlight featuring Kacper Klusek of Soba Labs on building enterprise RAG by fixing retrieval first: contextual chunking, BM25 hybrid search, and a reranker cut failed retrievals by 67% in Anthropic's evaluation, with LangSmith evals catching document drift.

介绍康奈尔大学与 ASAPP 的 DecisionRAG 论文：用决策模型（如 Jev 1.13）在文档树上导航取代向量相似度检索，实现 retriever-free RAG，在 Natural Questions 上 top-1 命中率达 65.4%，优于 embedding 检索加 8B reranker 的 48.0% 和 GPT-5 reranking 的 60.8%。

This article introduces Obelisk, an open-source system that indexes historical sessions from coding agents like Claude Code and Codex into SQLite for direct retrieval by agents. Its design is inspired by the paper "Is Complex Agentic Memory Always Needed?", which found that a simple RAW-RAG approach (retaining raw conversation history and retrieving on demand) outperformed complex agentic memory schemes across multiple memory benchmarks.

DecisionRank 使用 item-space on-policy distillation 将生成式教师模型的排序知识迁移到高效决策模型，在 TREC DL、BEIR、BRIGHT 等检索基准上达到最先进性能，同时推理速度显著提升（最高快 24.5 倍），1.7B 参数即可超越 27B 生成式教师。

A practitioner shares scores for AI agent retrieval/RAG tools from a rankings site they built, noting LlamaIndex leads at 7.0 while Pinecone and Ragas lag on release freshness, and asks production teams about their retrieval stacks and evaluation practices.

MemSteer, developed at Peking University and Microsoft Research Asia, converts user histories into automatically generated supervision tests for agent memory, and MemSteer-Opt uses the resulting evidence to optimize storage, retrieval, and presentation without hand-labeled data or fine-tuning, improving BM25 baselines on ATM-Bench and RHELM.

提出了ETAMem多视角记忆框架，通过解耦异构对话信息并构建Event、Affect、Time三个互补的层次化视图，提升长期对话智能体的记忆检索效果，在LoCoMo和LongMemEval-S上分别实现12.46%和7.96%的相对准确率提升。

TrawMem reframes LLM agent memory systems as task-oriented role-adaptive workspaces, organizing interaction histories into threads and selecting complementary memories by their task roles, achieving consistent macro-F1 gains of 2.23–3.14 points over representative memory systems on LoCoMo and MemGallery benchmarks.

A technical report compares taxonomy-label search with a two-stage embedding retrieval plus LLM filtering approach on a corpus of 7,354 randomized trials, finding the new method recovers 92% of systematic-review evidence versus 45% for label search, at a cost that scales with search depth.

FinLLM-KG 是一个模型无关的框架，将因果 LLM 与金融知识图谱结合，从 SEC 10-K 文件中注入公司级、时间对齐的知识以改进金融情绪分析与算法交易，同时提供可解释的预测依据。

This paper shows that retrieving the correct fact is insufficient for memory agents, as an obsolete competing statement harms accuracy more than unrelated evidence of the same size, evaluated on MemoryAgentBench's FactConsolidation task with Qwen2.5 and Qwen3 models.

该论文提出UMer框架，通过从用户历史社交痕迹中蒸馏个性化心理先验、进行量表引导的多查询检索以及粗到细的证据感知推理，实现基于社交媒体的抑郁严重程度估计，并提供可解释的证据依据。

FlexA-Mem 提出一种预算自适应的智能体记忆机制，将历史一次性写入不可变存储，并为每次调用动态渲染紧凑的时间线记忆加相关细节，在 AppWorld、LongMemEval 和 LoCoMo 上显著优于 A-Mem 等现有方法。

This paper proposes a new method for measuring whether LLM agents have internalized skills from public skill hubs, arguing that gain-based retirement rules fail to distinguish internalized skills from uncovered ones. It introduces a probe-based unaided scoring approach with cross-model validation and releases 1,251 skills with task data across four models.

A paper reports a systematic gap in modern text embedders: sentiment is highly decodable via linear probes but nearly absent from their cosine-similarity metric, and proposes SISE, a tiny supervised-contrastive head that fixes this on frozen encoders.

本文提出 VMem-Bench，一个两阶段诊断基准，系统研究长视频生成中视觉记忆（检索历史画面）对历史一致性与当前故事准确性的权衡影响，发现检索虽提升实体召回率却降低了精确度与禁用实体规避能力。

Policy Central 是一个已在警局部署的系统，通过两种架构（Retrieve-Judge 与 Screen-Confirm）自动检测政策文本段落之间的冲突、冗余与相似性，并在 302 对真实政策基准上报告了召回率与成本表现。