The agent successfully assembled a complete, evaluative explanation of TheoGrader's purpose, positioning, and pricing from the site and supporting sources, but discovered a critical mismatch: the site presents it as an operating commercial product with 200+ users, while GitHub reveals it as an unstarred academic final-year project with no evidence of real adoption. The site's core value proposition is clear and navigable, but maturity claims are unverified and potentially misleading.

• ›The homepage presents TheoGrader as an operating product with specific institutional pilots and user counts (200+ lecturers, 15 universities), but the official GitHub repository describes it as a final-year academic project with no commercial framing, no license, and zero stars or forks—a direct contradiction that the agent had to surface via external research rather than from site content.
• ›The site's stats counters all display '0' rather than the claimed figures, indicating they failed to load or are placeholders; the agent could not verify any usage metrics directly from the site.
• ›Critical site links (About, Research Paper, Contact, Privacy, Terms) are non-functional `#` placeholders, blocking access to claimed differentiators (NDPR compliance, data retention) and company context; only signup-linked `/terms` and `/privacy` paths exist but were not fetched.
• ›The signup page's incongruous heading ('Account Created / Upload your first rubric') and the unexplained 'NDPR Compliant' badge suggest the site's UX and messaging are incomplete or hastily assembled.
• ›The GitHub sequence diagram reveals that rubric extraction calls OpenAI's gpt-4o-mini, meaning student exam scripts and marking schemes are sent to a third-party LLM service—a data-handling fact not disclosed anywhere on the site, requiring external code inspection to discover.