There is no such thing as "THE mempool"
Every node has it's own mempool. Transactions that one node has in it's mempool, might not be in another node's mempool. There is nothing in the protocol that requires any nodes to keep any particular number of transactions in their mempool or to keep the transactions there for any particular amount of time. If you are checking two different sources, your are going to get the results based on how each mempool is configured.
I really thought there was a single mempool shared with all the nodes,so thank you for explaining that to me.