[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-179747-105":59,"doc-detail-179747-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","legal-llm-full-search-limitations-preincorporation-liability-founder-joint-liability","Legal LLM & Full Search Limitations - Preincorporation Liability Founder Joint Liability","","Analysis of legal large language model performance on pre-incorporation association scenarios, focusing on founder liability and how creditor claims change after establishment. Compares full-search prompting with introspective, agentic, and difficulty-aware strategies, identifying limitations such as introspection deficit, fragility in complex fact patterns, and failure to trigger verification. Evaluates search-trigger rates, correction success, and multiple LRAS/QL models across legal benchmarks, showing that targeted search and RL-based difficulty training improve accuracy while reducing hallucinated or overconfident reasoning.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/legal-llm-full-search-limitations-preincorporation-liability-founder-joint-liability/179747/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/legal-llm-full-search-limitations-preincorporation-liability-founder-joint-liability/179747.png","ImageObject",300,407,{"name":92,"@type":93},"Clementine","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-01","2026-09-02",true,{"@type":102,"interactionType":103,"userInteractionCount":39},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What legal issue is central to the document’s experiments?","Question",{"text":112,"@type":113},"The experiments focus on pre-incorporation association establishment and how liability and claims shift among founders, the association, and an individual who signed lease obligations.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"Why do full-search LLM approaches fail in some complex scenarios?",{"text":117,"@type":113},"They can suffer from an introspection deficit and fragility, failing to recognize which party should be liable and not triggering sufficient verification when reasoning becomes subtle.",{"name":119,"@type":110,"acceptedAnswer":120},"How does the difficulty-aware RL stage improve model performance?",{"text":121,"@type":113},"Difficulty-aware RL curates hard samples and trains models with difficulty-aware imitation/RL pipelines, which increases accuracy on legal benchmarks and raises the effectiveness of search-triggered corrections.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},179747,1788338093,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":39,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":46,"language":139,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":140,"faqs":141,"seo_title":142,"seo_description":67,"update_tm":129,"read_time":31,"preview_page_count":4},1374391974564,"https://ap-avatar.wpscdn.com/avatar/14000253aa45c000a9e?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779874745381141002","| Question: X, Y, and Z jointly contributed to establish \" ... Association\" (social group legal person) . During establishment, X et al. signed a 3-year shop [lease with](lease with ... in)[ ...](lease with ... in)[ in](lease with ... in)[ ](lease with ... in)[the Association](the Association)'[s](s) name.Y personally leased a high-grade printer from ... for the Association's establishment. Regarding the claims:\u003Cbr>A: ... lease claim extinguishes.\u003Cbr>B: Even if not established, ... claim lease debt from X et al.\u003Cbr>C: If not established, X et al. bear joint and several liability ...\u003Cbr>D: After Association established, ... can claim from either the Association or Y. |  |  |  |  |  |\n| --- | --- | --- | --- | --- | --- |\n| Legal LLM & Full Search Limitations |  |  |  |  |  |\n|  |  |  |  |  |  |\n| \u003Cbr>\u003Cbr> LLMs\u003Cbr>\u003Cbr>Question  Full Search LLM |  | [Introspection Deficit:](Introspection Deficit: ...lease)[ ...lease](Introspection Deficit: ...lease) signed in Association's name, debt transfers upon establishment, claim extinguishes...  |  |  |  |\n|  |  | Fragility in Complex Scenarios: ...Y signed personally, so only Y is liable, creditor can only claim from Y...  |  |  |  |\n|  |  |  |  |  |  |\n|  |  |  |  |  |  |\n| Legal Agentic Search |  |  |  |  |  |\n|  |  |  |  |  |  |\n| \u003Cbr>\u003Cthink> ... I need the specific Civil Code rules regarding \"founder liability\" ... \u003C/think>\u003Csearch>Civil Code Article 75 preincorporation liability founder joint liability\u003C/search>\u003Cbr>\u003Cbr>|  |  |  |  |  |\n\n| Category | Count (Percentage) |\n| --- | --- |\n| Total Base Model Failures | 705 (100%) |\n| Samples Where Search Was Triggered Samples Where Search Was Not Triggered | 202 (28.7%)\u003Cbr>503 (71.3%) |\n| \u003Cbr>Successfully Corrected by Search\u003Cbr>Search Success Rate (Corrected / Triggered) | \u003Cbr>58 (8.2% of Total)\u003Cbr>28.7% |\n\n|  |  |  |  |  |  |  |  |  |\n| --- | --- | --- | --- | --- | --- | --- | --- | --- |\n| 49.2 |  |  |  |  |  |  |  |  |\n| 39.0 |  |  |  |  |  |  |  |  |\n| 34.8 |  |  | 35.2 |  |  |  |  |  |\n|  |  |  |  |  |  |  |  |  |\n|  |  |  |  |  |  |  |  |  |\n|  |  |  |  |  |  |  |  |  |\n\n| Introspective Data Synthesis\u003Cbr>Introspective Prompt\u003Cbr>LLM |  |  |\n| --- | --- | --- |\n|  | Legal Question: xxx\u003Cbr>\u003Cthink> xxx \u003C/think>\u003Cbr>\u003Csearch> xxx \u003C/search>\u003Cinformation> xxx \u003C/information>\u003Cthink> xxx \u003C/think>\u003Cbr>\u003Canswer> xxx \u003C/answer> |  |\n\n| Stage 2: Difficulty-aware RL |\n| --- |\n| \u003Cbr>Data Curation |\n| Candidate Set\u003Cbr>\u003Cbr>\u003Cbr>Difficulty RL Datasets\u003Cbr>(Hard samples) |\n\n| |  |\n| --- | --- |\n| Base Model Introspective Imitation Learning | LRAS SFT Model Difficulty RL LRAS RL Model |\n\n| Model | Parameters | LexEval | LawBench | Unilaw-R1-Eval | DiscLaw | Avg.(%) |\n| --- | --- | --- | --- | --- | --- | --- |\n| Legal LLMs |  |  |  |  |  |  |\n| Fuzi-7B | 7B | 13.56 | 41.22 | 6.50 | 17.55 | 19.71 |\n| LawLLM-7B | 7B | 49.72 | 77.87 | 29.12 | 48.25 | 51.24 |\n| DiscLaw-13B | 13B | 35.07 | 59.49 | 22.38 | 33.11 | 37.51 |\n| Legal∆-3B | 3B | 33.58 | 71.39 | 26.50 | 51.56 | 45.76 |\n| Legal∆-7B | 7B | 48.02 | 75.78 | 37.88 | 45.75 | 51.86 |\n| Legal∆-14B | 14B | 49.74 | 80.42 | 45.88 | 52.14 | 57.05 |\n| Qwen3-4B |  |  |  |  |  |  |\n| Base | 4B | 38.96 | 61.67 | 30.50 | 49.47 | 45.15 |\n| LRAS-SFT | 4B | 51.37 | 72.38 | 38.25 | 61.64 | 55.91↑23.8% |\n| LRAS-RL | 4B | 51.40 | 76.83 | 44.38 | 65.72 | 59.58↑32.0% |\n| Qwen3-8B |  |  |  |  |  |  |\n| Base | 8B | 49.22 | 71.96 | 39.50 | 56.93 | 54.40 |\n| LRAS-SFT | 8B | 55.29 | 75.94 | 49.38 | 67.18 | 61.95↑13.9% |\n| LRAS-RL | 8B | 57.44 | 78.76 | 51.00 | 73.23 | 65.11↑19.7% |\n| Qwen3-14B |  |  |  |  |  |  |\n| Base | 14B | 52.28 | 73.88 | 43.00 | 66.59 | 58.94 |\n| LRAS-SFT | 14B | 57.07 | 78.76 | 49.38 | 71.20 | 63.80↑8.2% |\n| LRAS-RL | 14B | 59.84 | 80.10 | 54.37 | 75.66 | 67.49↑14.5% |\n\n|  Direct Reasoning |\n| --- |\n| 1. Memory Recalling:\u003Cbr>\"Recalling Constitution Art 89, 93... \" |\n| 2. Knowledge Hallucination:\u003Cbr>\"The CMC is a state organ, so it must report work like the State Council.“\u003Cbr>Result: Fails to recall that CMC Chairm","cbCaipUAoaClGS3g","https://ap.wps.com/l/cbCaipUAoaClGS3g","pdf",989570,"English","# Legal LLM & Full Search Limitations\n## LLMs: Full Search and Introspection Deficit\n## Legal Agentic Search\n## Stage 2: Difficulty-aware RL\n### Data Curation\n### Difficulty RL Datasets\n## Model Evaluation Results\n### LexEval, LawBench, Unilaw-R1-Eval, DiscLaw\n## Direct Reasoning Failure Modes","[{\"question\":\"What legal issue is central to the document’s experiments?\",\"answer\":\"The experiments focus on pre-incorporation association establishment and how liability and claims shift among founders, the association, and an individual who signed lease obligations.\"},{\"question\":\"Why do full-search LLM approaches fail in some complex scenarios?\",\"answer\":\"They can suffer from an introspection deficit and fragility, failing to recognize which party should be liable and not triggering sufficient verification when reasoning becomes subtle.\"},{\"question\":\"How does the difficulty-aware RL stage improve model performance?\",\"answer\":\"Difficulty-aware RL curates hard samples and trains models with difficulty-aware imitation/RL pipelines, which increases accuracy on legal benchmarks and raises the effectiveness of search-triggered corrections.\"}]","Legal LLM & Full Search Limitations - Preincorporation Liability Founder Joint Liability | PDF"]