Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for improving search results. In one aspect, a method includes receiving documents responsive to a query, each document having an associated score indicative of the documents relevance to the query. The method includes determining, for a plurality of the associated scores, a distribution of the scores along a dimension that measures trustworthiness of the scores. The method also includes reducing each of one or more of the plurality of associated scores by an amount determined based on the scores respective trustworthiness and properties of the distribution, and ranking the documents to account for the reduced scores.
