The agent successfully assembled a comprehensive explanation of Arcade by combining official site content (homepage, pricing page, llms.txt, openapi.json) with third-party review sources. The site was moderately navigable — it publishes machine-readable artifacts (llms.txt, openapi.json) and a pricing page, but those resources contained incomplete or JavaScript-heavy pricing details, forcing the agent to supplement with external sources and make inferences about confusing pricing tiers.

• ›Steps [1], [6], [7] returned authoritative product positioning and capabilities from the homepage and llms.txt, establishing what Arcade does (MCP runtime for agent authorization, tool execution, governance). Step [9] (openapi.json) provided technical API context but less business-facing information.
• ›Steps [3] and [13] (pricing page) were cited but contained minimal actionable content — the agent noted the page 'loads mostly JavaScript' and had to piece together the actual pricing structure (Hobby/Growth/Enterprise tiers with per-execution costs) from third-party sources (step [15] search results and step [14] xpay.sh aggregator). The official pricing page failed to surface structured pricing data.
• ›Step [16] (Merge.dev alternatives article) was explicitly cited and provided competitive positioning, but the core differentiation claim (authorization-first, Okta/Snowflake pedigree) came from step [6]'s llms.txt, not from the site's own marketing materials.
• ›The agent encountered a 404 at /docs (step [4]), indicating incomplete site structure. It compensated by discovering llms.txt and llms-full.txt probes via prior knowledge, not via links from the homepage, suggesting these artifacts are not prominently surfaced in the information architecture.
• ›The final answer's confusions (vague 'standard' vs. 'pro' execution distinction, missing self-hosted pricing, absent tool catalog) directly reflect gaps in what the official site published — the agent could not explain what it was not told.