The agent successfully understood Supabase through web search and third-party resources rather than the official site itself. The supabase.com domain was heavily JavaScript-rendered and returned unreadable HTML, forcing the agent to rely on external comparison articles and guides (Jetadmin, DevArt, Devart, MindStudio) and its own prior knowledge to construct a complete answer about what Supabase does, who it's for, how it's priced, and how it differs from Firebase. The official site proved difficult to navigate programmatically but the agent assembled sufficient information to satisfy the open-ended discovery task.

• ›The official supabase.com pages (steps 1–7, 12) returned only client-side rendered Next.js HTML with no readable content. All actionable information came from web search results and external third-party articles, not the domain itself.
• ›Step 9 (search query about features and pricing) and step 13 (Jetadmin pricing guide) were the primary sources for pricing tiers and structure; step 10 (Supabase vs. Firebase search) surfaced the direct comparison resource. These searches succeeded where direct site fetches failed.
• ›The agent correctly identified and cited external sources in its final response, demonstrating awareness that official site content was unavailable. Step 13's third-party pricing guide became a critical substitute for missing Supabase.com/pricing rendered content.
• ›The site's JavaScript-heavy architecture created a significant barrier: pricing details, comparison tables, and feature lists that should be publicly available are client-rendered and inaccessible via HTTP fetch alone. An agent without search fallback would have failed entirely.
• ›The agent used prior knowledge (25%) to fill gaps on context about PostgreSQL, vendor lock-in, and general backend platform positioning—suggesting the site does not make these differentiators explicit enough to extract programmatically.