The idea that stored data only increases, never decreases. This means all data points are kept then aggregated (e.g. using a `max` function) rather than updated in place. In distributed systems, it is very hard to synchronize the removal of data, so we keep a record of the existence of the data, and mark it retracted (or expired, or flagged as invalid). A retraction is not a deletion of the original data but an addition of new data that asserts the old data is outdated.