In Uncertain Domains, Premature Evidence Can Mislead More Than Conviction

**What assumption do I disagree with?** The original assumes that moving "from opinion toward evidence whenever possible" is an improvement by definition. In stable, well-understood domains — medicine, established engineering, accounting — this is correct. But in domains defined by radical uncertainty, the evidence available at decision time is systematically misleading. This is exactly what Clayton Christensen documented in The Innovator's Dilemma: well-managed incumbent companies failed not because they ignored evidence, but because they followed it too carefully. Their customers gave them accurate feedback — the new disruptive technology performed poorly on every established metric — and the companies rationally responded by not investing in it. The evidence was real. The problem was that it measured the wrong thing at the wrong stage. In early-stage startups, the same pattern repeats constantly. Initial customer feedback tends to describe confusion with a new paradigm, not underlying demand. Early retention metrics measure whether the current UI is clear, not whether the problem is worth solving. A knowledge norm that rewards "evidence over opinion" in these contexts pushes founders to over-index on noise at exactly the moment they need the judgment to hold conviction. **What conclusion is different?** The original concludes that knowledge systems should systematically encourage participants to move from opinion toward evidence. The alternative view is that the relevant question is not "which is better — evidence or opinion?" but "how reliable is the specific evidence available in this context?" A founder with three years of domain experience who has spoken with 500 customers, holds a conviction about an unmet need, and can articulate why the current market evidence is measuring the wrong proxy — that founder is in a better epistemic position than a survey of 50 early adopters showing 60% churn. Systematizing the norm that evidence beats conviction rewards whoever has fast access to data, not whoever has earned the right to hold a strong view. The most valuable intellectual contributions often come from people who maintained conviction through periods of adverse evidence precisely because they understood why that evidence was misleading. None of this is an argument for baseless assertion. The original is right that unsupported opinion is less useful than evidence-backed claims in domains where evidence is reliable. But the claim as stated is universal, and it fails on contact with the conditions where the stakes are highest: frontier domains where the evidence is sparse, lagged, or definitionally unable to signal what matters most. A more precise formulation would be: evidence quality relative to conviction quality determines which is more reliable. A knowledge system worth building should help people develop and articulate high-quality conviction — including explaining why the available evidence is insufficient — not simply reward whoever has assembled the most data points.