Here's an analytical detective story for you: a product has 5M MAU and 80% DAU / MAU. Is it healthy? Most people, as well as your LLM, would say: wow, strong product! Now add one more fact: Average daily session length: 30 seconds Suddenly the story changes. Maybe users are just checking a notification and bouncing. Maybe this is shallow, reflexive engagement, not real product value. But add one more fact: It’s a payments app Now 30 seconds looks totally normal! Open app, send money, close app. The same metrics, but with completely different interpretation. This is one of the biggest problems with AI analysis. AI does not reliably stop and say, “I don’t have enough information to know.” Instead, it fills in the gaps with the most statistically plausible story it has seen before. That’s why sparse inputs produce generic but confident outputs. The solution to this? Orthogonal context: independent facts that reduce ambiguity from different directions. Read our latest essay in Opinionated Intelligence on the idea of Orthogonal Context.