This concept is detailed in a GitHub issue (https://github.com/bitcoin/bitcoin/issues/28699) and involves maintaining a bounded memory list of all transactions that have entered one's mempool at least once, provided they meet the current minimum fee requirements.

The proposal recognizes that not all full-nodes can afford extensive storage, particularly when facing attackers with moderate liquidity capabilities. In such cases, certain nodes might engage in periodic altruistic rebroadcasting to support transaction relay. However, this could lead to a tiered hierarchy of full-nodes, potentially making the network topology more susceptible to sybil attacks or the malicious squatting of scarce inbound connection slots.

A further concern is the default rate of altruistic rebroadcasting, which could be manipulated in amplification attacks. Attackers might exploit discrepancies between the rebroadcast traffic's fee rate and the dynamic minimum fee rate of the majority of network mempools, causing unnecessary bandwidth wastage.

Moreover, in scenarios involving medium-liquidity or high-liquidity attackers, current mitigation efforts may prove inadequate. Such attackers could generate an unbounded number of replacement transactions from a limited set of unspent transaction outputs (UTXOs), leading to concurrent spending issues. In multi-party time-sensitive protocols, honest counterparties' transactions with higher fee rates could be overshadowed by lower fee rate replacement transactions from malicious actors, thereby monopolizing the additional replacement cache storage and undermining the protocol's integrity.