The news coverage of events often contains not one but multiple incompatible
accounts of what happened. We develop a query-based system that extracts
compatible sets of events (scenarios) from such data, formulated as one-class
clustering. Our system incrementally evaluates each event'