The agent successfully assembled a comprehensive explanation of Lacuna by combining information from the site itself (homepage, about, explore, MCP pages) with the published research paper on ArXiv and the parent company's homepage. However, the site itself was poorly structured for answering key questions—pricing and target audience were entirely missing from on-site content, forcing the agent to infer 'free and open' from licensing metadata and to rely on external sources (ArXiv paper, secondary articles) for competitive differentiation and use-case clarity. The site is React-rendered with sparse text content and minimal navigation, making it difficult for agents to extract answers without supplementary research.

• ›Step [1] (homepage) returned metadata showing '753,265 papers across 38,150 research directions' but provided no explanation of what the tool does, who it's for, or how to use it—just raw statistics. The agent had to infer the purpose from the page title ('Open ML Research Commons') and description fragment.
• ›Step [2] (About page) promised to explain the product but the HTML response was truncated and appears to be a JavaScript-rendered SPA with minimal static content, making it unreadable in the fetched form. The agent could not extract actionable content from it.
• ›Steps [9] and [10] (ArXiv paper on Lacuna) were critical to the final answer and were explicitly cited as sources. The paper provided technical comparisons to OpenScholar, explained the research directions concept, and clarified that the tool uses LLMs to generate structured metadata—information completely absent from the main website.
• ›Step [11] (Tiptree Docs on literature reviews) was cited as a source but provided limited insight; it focused on Althea (a sibling product) rather than Lacuna's core offering.
• ›Pricing page (step [3]) returned 404, confirming the site does not publish pricing. The agent inferred 'free and open' from the CC-BY 4.0 license metadata on the homepage, but this inference was never validated on-site.
• ›The site has no explicit messaging about target audience (researchers, students, R&D teams), no use-case examples, and no comparison to competitors—all of this came from the ArXiv paper or agent inference, not from on-site content.
• ›Navigation is minimal (only MCP and About links visible from homepage). The Explore section exists but is not explained; users are dropped into a search interface with no onboarding.
• ›The site is agent-hostile: React-rendered with low static text content, no structured data (schema.org, JSON-LD), no explicit calls-to-action, and key pages (About, pricing) either missing or inaccessible in fetched form.