The agent gathered a reasonably complete overview of durum.ai's core product, audience, and pricing from two fetched pages (homepage and pricing), but encountered deliberate and accidental gaps that prevented full clarity: a 404 comparisons page, unstated AI credit allowances despite conflicting claims, ambiguous spend measurement rules, and undefined terms like 'Studio visuals'. The site was partially navigable and content was mostly present but fragmented and incomplete.

• ›The homepage and pricing page contained the core facts (what durum.ai does, who it's for, pricing bands, AI credit mechanics), but critical details were missing or contradictory — e.g., homepage claims 7,000/20,000/50,000 credits per plan, but the pricing page says credit allowances are 'not stated,' forcing the agent to flag a content conflict rather than resolve it.
• ›A 'comparaisons d'outils' link was visible in the navigation but the guessed URL (/fr/comparaisons) returned 404, preventing the agent from directly verifying the site's own competitive claims. The agent disclosed this gap rather than filling it from model memory.
• ›Ad spend measurement and billing mechanics were underspecified: the site doesn't clarify whether spend is calendar month or rolling 30 days, which platforms count, or how overages are handled across plan boundaries. The pricing page explicitly noted 'not stated' for several of these.
• ›The Durum AI feature status was ambiguous: labeled 'early access' on the homepage but also shown in demos as operational, leaving unclear what customers actually have access to. The agent flagged this as genuine site confusion rather than agent error.
• ›The agent correctly resisted filling gaps with model memory — it cited missing features (Shopify/Stripe role, 'Pilotage automatique' definition, privacy/retention details, refund terms) as 'confusing or missing' rather than inventing explanations, demonstrating appropriate epistemic caution.