The agent successfully assembled a comprehensive understanding of Salus by visiting the homepage, pricing, features, docs, and audience-specific pages, supplemented by the llms.txt artifact which provided structured positioning. The site is reasonably navigable for discovery, with clear audience pages and a docs section, though key competitive differentiation and pricing tier feature gaps required inference rather than explicit documentation.

• ›Step [8] (llms.txt) was the most information-dense fetch, providing structured positioning on what Salus is, its MCP-native agent governance model, and core value propositions in machine-readable form—this artifact made the difference between scattered marketing claims and coherent product understanding.
• ›Steps [1, 2, 3] (homepage, pricing, features) and steps [10–13] (audience-specific pages for agents, builders, engineering, enterprise) all returned content, but the agent had to stitch together a coherent story across multiple pages; no single page answered 'what is this and who is it for?' holistically.
• ›Pricing structure was extractable from step [2], but the agent notes the response was truncated and feature-parity across tiers (what governance features are in Pro vs. Enterprise) was unclear—the site exposes pricing but not the detailed feature matrix.
• ›The agent guessed at URLs like /compare, /why-salus, /about, and /blog (steps [5, 7, 9]) based on prior knowledge; most returned 404, indicating the site has a minimal nav structure. The agent had to rely on 46% prior knowledge and the /llms.txt probe to fill gaps.
• ›Competitive positioning was inferred from the MCP-native agent governance angle and data sovereignty emphasis (BYOC), but the site does not name or directly compare against alternatives (Vercel, Railway, Fly.io, etc.), limiting direct competitive clarity.
• ›The llms.txt artifact was discovered via step [8] following a dead link (step [7] /about → 404), but the agent recognized it as a high-value probe and fetched it—this shows the site does publish structured product metadata, but discovery of that metadata is not obvious from normal navigation.