We discuss a simple data structure we designed for an approximate membership filter with probabilistic time decay. Our data structure needs to remember newer data with a higher probability and older data with a lower probability. In addition, we are going to have a large number of filters in the system, so a low memory overhead is required. Lastly filters are serialized/deserialized frequently, and they need to be done fast. Therefore, we looked for a different solution than LRU, and we came up with the data structure in this blog.