The agent could only partially understand marcode.net because the site itself is poorly discoverable and lacks standard documentation. Step [1] fetched the homepage HTML but it was JavaScript-heavy and truncated; step [9] accessed external GitBook documentation for a Sketch2React project. The agent assembled most of its answer from web search snippets and prior knowledge rather than from content the site directly published, and explicitly called out missing information about pricing, project details, and clear positioning.

• ›Step [1] (homepage fetch) returned only a Next.js-rendered stub with image preload tags and truncated HTML — no readable content about what the site does, who built it, or how projects are organized. The agent could not extract answers from the site itself.
• ›Step [9] (GitBook fetch) was the only substantive external documentation found, but it was for Sketch2React/marcode (a separate tool), not the main site's projects. This highlights the site's lack of self-contained docs.
• ›The agent filled critical gaps (what marcode.net is, who Marco De Nichilo is, what projects exist) using web search results [3], [10], [11], [14] and cited LinkedIn as a primary source — signals that the site itself does not publish this metadata in an agent-readable format.
• ›The agent explicitly flagged missing information: no /pricing endpoint (404 in step [6]), no accessible /about page (404 in step [5]), no project details or GitHub links discoverable from the homepage, and confusion between marcode.net, marcode.ai, and marcode.org.
• ›The site is a portfolio/project showcase but does not use structured data (Schema.org, metadata tags, or text content) to advertise its purpose, projects, pricing, or differentiation — all hallmarks of poor agent-readiness.