Searching in the Dark and Learning Where to Look

A searcher knows little about how items' observable attributes map to qualities but wants to avoid missing good discoveries. Knowing only that items with similar attributes have similar qualities, she sequentially decides where to look next or stops searching to take her best discovery. We characterize worst-case optimal search procedures. The relationship between search effort and complexity is cliff-shaped and reflects choice overload. The search path is funnel-shaped if there is a `sweet spot' in attribute space.