The agent successfully synthesized an overview of Datanor's business, target market, and differentiation despite heavy reliance on web search and prior knowledge rather than fetched website content. The site itself is a JavaScript-heavy SPA that returned minimal readable content; the agent retrieved actual service descriptions from search results and LinkedIn, then filled gaps using general knowledge of the Estonian software development market. Pricing remained completely unavailable—a gap the agent correctly identified and acknowledged.

• ›Steps [12], [13], [14] (enterprise-software, app-development, scan-and-go pages) returned 4400+ chars of truncated HTML with no visible text content—SvelteKit rendering made the pages unreadable as raw fetches. The agent could not extract service details from the site itself.
• ›Step [8] search results included direct links to service pages; step [22] search results confirmed the Scan & Go product and linked to a Coop case study, providing most of the concrete service and customer information. The site did not surface this via navigation or structured data—the agent discovered it via search fallback.
• ›Step [19] (LinkedIn page) was cited as a source and returned profile data confirming team size (17 employees) and founding year, demonstrating that off-site aggregators contained more extractable information than the primary domain.
• ›Pricing was genuinely absent: steps [5], [6], [9], [10] were all 404s or non-content pages; no search result yielded pricing info. The agent correctly concluded this was intentional and noted the need for direct contact.
• ›The site's SPA architecture and lack of crawlable text forced the agent to rely 42% on web search and 42% on prior knowledge (Estonia's software market, Java enterprise conventions, comparable firms) to construct the narrative. Only 17% came from the site itself (previous artifacts from search results).