{"slug":"reasoning-question","label":"Reasoning Question","item_count":3,"day_count":3,"source_count":2,"first_seen":"2026-10-23T09:50:37+00:00","last_updated":"2026-10-27T07:14:59+00:00","generated_at":"2026-10-29T00:13:38.528542+00:00","sources":["arxiv_agent_systems_research","arxiv_llm_reliability"],"days":[{"date":"2026-10-23","items":[{"title":"Cycle-of-Thought: Self-Evaluative Iterative Reasoning for Table Visual Question Answering in Document Images","url":"http://arxiv.org/abs/2610.32423v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Recent advances in Vision-Language Models (VLMs) have significantly improved Visual Question Answering (VQA) performance, yet their reasoning over structured content such as tables remains inconsistent, often plagued...","why_it_matters":"Matches feed focus: evaluation.","sid":"1846766ba5a6742d","published":"2026-10-23T09:50:37+00:00","editor_note":"Self-evaluative iteration for table reasoning in document images."}]},{"date":"2026-10-26","items":[{"title":"Evolve on Graph: Self-Evolving Knowledge Graph Reasoning without Question Annotations","url":"http://arxiv.org/abs/2610.35598v1","source":"arxiv_llm_reliability","type":"paper","summary_1line":"Knowledge graphs (KGs) provide reliable external knowledge for multi-hop factual reasoning with large language models (LLMs), yet most KGs in real-world domains such as finance and industry lack question--answer (QA)...","why_it_matters":"Matches feed focus: agent, eval.","sid":"26fc92a7573c4011","published":"2026-10-26T15:23:04+00:00","editor_note":"Removes the need for labeled questions when reasoning over knowledge graphs."}]},{"date":"2026-10-27","items":[{"title":"DocSkill: Skill-Integrated Agentic Reasoning for Multilingual Multi-Hop Document Question Answering","url":"http://arxiv.org/abs/2610.36759v1","source":"arxiv_agent_systems_research","type":"paper","summary_1line":"Multilingual multi-hop document question answering (MMHop-DocQA) requires locating and integrating evidence across complex layouts, document pages, and diverse languages. Existing approaches include single-pass proces...","why_it_matters":"Matches feed focus: agentic, eval.","sid":"334116357c766a21","published":"2026-10-27T07:14:59+00:00","editor_note":"Adds an agentic, skill-based approach for multilingual multi-hop documents."}]}],"editorial":{"tldr":"Three arXiv papers in five days attack question answering over structured content: tables in document images, knowledge graphs without labeled questions, and multilingual multi-hop documents. All three add structure or self-evaluation around the model instead of relying on a single pass.","stale":false,"whats_new":"DocSkill (Oct 27) applies skill-integrated agentic reasoning to multilingual multi-hop document QA, where evidence spans complex layouts, pages and languages.","why_it_matters":"Document and KG question answering is a common production RAG workload, and these papers move reliability gains into orchestration (iterative self-checks, skills, self-generated training signal) you can adopt without a new base model.","take_for_builders":"If you run document or table QA in production, test an iterative self-check pass on your own failure cases before adopting any of these; none shows production-grade evidence yet.","status":{"state":"Developing","tone":"rising","changed":"2026-10-27","detail":"Research-stage papers only; no released code, benchmarks or production results are cited in this thread."},"beats":[{"kicker":"TABLES","tone":"launch","headline":"Cycle-of-Thought adds self-evaluation to table VQA","summary":"Iterative self-checking targets inconsistent VLM reasoning over tables in document images.","sids":["1846766ba5a6742d"]},{"kicker":"GRAPHS","tone":"rising","headline":"Evolve on Graph trains KG reasoning without question annotations","summary":"Aimed at finance and industry KGs that lack QA pairs.","sids":["26fc92a7573c4011"]},{"kicker":"NOW","tone":"now","headline":"DocSkill brings skills to multilingual multi-hop document QA","sids":["334116357c766a21"]}],"open_questions":["Do any of the three papers release code or evaluation harnesses that others can reproduce?","Does DocSkill report gains over single-pass baselines on multilingual document benchmarks?"],"generated_at":"2026-10-29T00:13:08+00:00"}}