The agent successfully assembled a comprehensive explanation of Supernova.io's purpose, target users, pricing, and competitive differentiation, but relied heavily on web search and prior knowledge rather than direct extraction from the site itself. The supernova.io domain uses client-side rendering that made direct content extraction difficult ([1], [3]), forcing the agent to pivot to web search results and external documentation. The agent retrieved enough information from search snippets, documentation pages, and prior knowledge to satisfy the open-ended discovery task, but the site's own marketing pages were not effectively machine-readable.

• ›The homepage and pricing page ([1], [3]) returned client-side rendered HTML that contained minimal extractable text, requiring the agent to abandon direct site scraping and rely on web search as its primary source.
• ›Web search results ([5], [11], [13], [17], [19], [21]) provided the actual content-bearing material—the search snippets themselves contained positioning language, feature descriptions, and pricing tier names that the agent used to construct its answer. The agent never successfully fetched SaaSWorthy or Capterra ([7], [9]) due to Cloudflare blocks.
• ›The agent discovered key differentiators (AI-first architecture, MCP server, continuous deployment via Supernova Relay) through targeted searches and the learn.supernova.io documentation fetch ([15]), not from the main marketing site, suggesting the site splits positioning content across multiple domains.
• ›The agent explicitly cited 9 sources in its final response, but only [15] was a direct fetch; the rest were inferred from search result titles or represented prior knowledge synthesized from the search results. This indicates the site's core messaging is not concentrated in a single, easily-parseable landing page.
• ›Pricing specifics (Free, Pro at $20–$35/seat/month, Enterprise custom) came from search result snippets and external references rather than direct site extraction, suggesting pricing page content was not machine-readable from the fetch in [3].
• ›The agent flagged genuine gaps: exact builder vs. full seat definitions, free plan AI credit limits, MCP server data structure details, and migration guidance were not cleanly surfaced in any single source it could reach, indicating the site spreads this information across marketing, docs, and developer portals without a single source of truth.