[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85467-en":3,"doc-seo-85467-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},85467,549758146520,"Patrick","https://ap-avatar.wpscdn.com/avatar/80002397d8c0411e94?_k=1775819394049821470",8,"Research & Report","Mini Amusement Parks (MAPs): A Testbed for Modelling Business Decisions","Despite rapid progress in artificial intelligence, current systems struggle with interconnected challenges that shape real-world decision making. Business management requires open-ended optimization, active learning from sparse experience, planning over long stochastic horizons, and reasoning over spatial information. No existing human–AI benchmark unifies these demands in a grounded setting. Mini Amusement Parks (MAPs) evaluates agent environment modelling, long-term uncertainty anticipation, and strategic operations, with expert humans outperforming agents by 11.4× (easy) and 15.3× (medium). Weaknesses persist in long-horizon planning, sample-efficient learning, spatial reasoning, and uncertainty modelling.","arXiv :2511 . 15830v2 [ cs .AI] 10 Jul 2026  \nMini Amusement Parks (MAPs): A Testbed for Modelling Business Decisions  \nStéphane Aroca-Ouellette*, 1 Ian Berlot-Attwell*, 1, 2, 3 Panagiotis Lymperopoulos1 Abhiramon Rajasekharan1 Tongqi Zhu1 Herin Kang1 Kaheer Suleman1  \nSam Pasupalak1  \n* Equal Contribution, [1](1 Skyfall.ai)[ Skyfall.ai](1 Skyfall.ai),2 University of Toronto,3 Vector Institute  \nAbstract  \nDespite rapid progress in artificial intelligence, current systems struggle with the interconnected challenges that define real-world decision making. Practical domains such as business management require open-ended optimization, actively learning environment dynamics from sparse experience, planning overlong horizons in stochastic settings, and reasoning over spatial information. Yet no existing human–AI benchmarks assess how well agents integrate these challenges in a grounded decision-making context. To this end, we introduce Mini Amusement Parks (MAPS), an amusement-park simulator designed to evaluate an agent’s ability to model its environment, anticipate long-term consequences under uncertainty, and strategically operate a complex business. We provide expert human performance and a comprehensive evaluation of state-of-the-art agents, finding experts outperform these systems by 11.4× on easy mode and 15.3× on medium mode. Our analysis reveals persistent weaknesses in long-horizon planning, sample-efficient learning, spatial reasoning, and modelling uncertainty. By unifying these challenges within a single environment, MAPS offers a new foundation for benchmarking agents capable of adaptable decision making. Code:  \n[https://github.com/Skyfall-Research/MAPs](https://github.com/Skyfall-Research/MAPs)  \n1 Introduction  \nAI systems have achieved or surpassed human performance on a wide range of tasks and exams [40, 28, 25, 29] . Yet these successes are consistently in environments with narrow objectivesand minimal uncertainty. This differs from many tasks in human domains: for example, running a business. Here, success hinges on coordinating interdependent decisions—allocating staff, developing infrastructure, investing in R&D, and conducting market research—while simultaneously reasoning over long horizons about noisy, partially observable customer behavior shaped by multiple, often multi-modal, interacting factors. Moreover, business owners must actively and efficiently learn how to navigate these challenges, or risk bankruptcy.  \nEach of these dimensions, while frequently navigated by humans, presents a major hurdle for stateof-the-art (SotA) AI systems. (1) When applied directly, large language models (LLMs) perform poorly on long-horizon planning and sequential reasoning tasks [43, 5], often exhibiting myopic optimization for immediate concerns [48, 31] . Mitigation methods, such as tree search or iterative reasoning [53], quickly become expensive and are limited by the model’s often inadequate evaluation of nodes [10] . (2) AI systems remain inefficient learners, struggling to infer patterns from a handful of examples [12] . In contrast, humans actively construct and test hypotheses to accelerate learning [19]  \nPreprint.  \nTable 1: Comparison of benchmarks that report human performance. ✗ = absence ▲ = partial presence ✓ = presence.  \n\n|  | CH1 CH2 CH3 CH4 CH5 |  |  |  |  | Features |  |\n| --- | --- | --- | --- | --- | --- | --- | --- |\n| Environment | Open-Ended\u003Cbr>Objective | Long-Horizon\u003Cbr>Planning | Active WM\u003Cbr>Learning | Spatial\u003Cbr>Reasoning | Stochastic\u003Cbr>Transitions | Online Variable\u003Cbr>Leaderboard Difficulty | Playable\u003Cbr>Online |\n| ARC-AGI2 AutumnBench Crafter DiscoveryWorld WebShop\u003Cbr>WebArena OSWorld VendingBench MAPS | ✗ ✗ ▲ ✓ ✗\u003Cbr>✗ ▲ ✓ ✓ ✓\u003Cbr>✗ ✓ ▲ ✓ ✗\u003Cbr>✗ ✓ ▲ ✓ ✗\u003Cbr>✗ ✓ ▲ ✓ ✗\u003Cbr>✗ ✓ ▲ ✓ ✗\u003Cbr>✗ ✓ ▲ ✓ ✗\u003Cbr>✓ ✓ ▲ ✗ ✓\u003Cbr>✓ ✓ ✓ ✓ ✓ |  |  |  |  | ✓ ✓ ✓\u003Cbr>✗ ✗ ✓\u003Cbr>✗ ✗ ✗\u003Cbr>✗ ✓ ✗\u003Cbr>✗ ✗ ✗\u003Cbr>✓ ✗ ✗\u003Cbr>✓ ✗ ✗\u003Cbr>✓ ✗ ✗\u003Cbr>✓ ✓ ✓ |  |\n\na capacity largely unexplored in current AI research (3) Existing models lag","cbCaidP0NLCpVxOH","https://ap.wps.com/l/cbCaidP0NLCpVxOH","pdf",2547613,1,46,"English","en",105,"# Abstract\n# Introduction\n## Human–AI benchmark gaps\n## MAPS as a unified testbed","[{\"question\":\"What problems does MAPS aim to benchmark for AI decision-making agents?\",\"answer\":\"MAPS benchmarks agents on environment modelling, long-term consequence anticipation under uncertainty, active learning from limited experience, and spatial reasoning for strategic operations in a business-like simulator.\"},{\"question\":\"How does MAPS differ from previous human-baseline benchmarks?\",\"answer\":\"Prior benchmarks typically cover only subsets: some focus on long-horizon planning with simplified dynamics, others test abstract few-shot generalization, and some evaluate practical tasks but lack unified uncertainty optimization or active learning.\"},{\"question\":\"What do the reported human vs. agent results show?\",\"answer\":\"Expert human performance surpasses state-of-the-art agents by 11.4× on easy mode and 15.3× on medium mode, indicating persistent weaknesses in long-horizon planning, sample-efficient learning, spatial reasoning, and modelling uncertainty.\"}]",1784203774,116,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"mini-amusement-parks-maps-a-testbed-for-modelling-business-decisions","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/mini-amusement-parks-maps-a-testbed-for-modelling-business-decisions/85467/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problems does MAPS aim to benchmark for AI decision-making agents?","Question",{"text":75,"@type":76},"MAPS benchmarks agents on environment modelling, long-term consequence anticipation under uncertainty, active learning from limited experience, and spatial reasoning for strategic operations in a business-like simulator.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does MAPS differ from previous human-baseline benchmarks?",{"text":80,"@type":76},"Prior benchmarks typically cover only subsets: some focus on long-horizon planning with simplified dynamics, others test abstract few-shot generalization, and some evaluate practical tasks but lack unified uncertainty optimization or active learning.",{"name":82,"@type":73,"acceptedAnswer":83},"What do the reported human vs. agent results show?",{"text":84,"@type":76},"Expert human performance surpasses state-of-the-art agents by 11.4× on easy mode and 15.3× on medium mode, indicating persistent weaknesses in long-horizon planning, sample-efficient learning, spatial reasoning, and modelling uncertainty.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]