A massively parallel real-time computing system receives input data events across many compute nodes, each with a processing algorithm in its processing pipeline. An Event Manager is placed before the algorithm processing pipelines, receives metadata about each incoming event, and collects and organizes it in a database. A fast histogram compares the metadata about each event to that of all the other events, in a processing interval. For sufficiently matching metadata, the events are forwarded to the processing nodes as "regular" events for processing. If the metadata for a processing interval does not match sufficiently, the histogram decides which events are the "correct" events and which events are "incorrect". The "correct" events are sent on for processing and the "incorrect" events are combined with the "correct" metadata and sent back to the processing nodes to supplement or modify their incoming data to match the other nodes' expectations.