The consensus rules specify what a node must accept in a block. These rules must match perfectly across all node implementations. If a block or a transaction in a block are evaluated differently by some nodes, the network would experience a consensus failure as some nodes accept the block while others fork off.

A node’s mempool policy defines what the node will relay and consider for the block templates it produces. We use mempool policy to e.g. discourage transactions that are expensive to validate or might trigger security issues for some nodes, and the premature use of upgrade hooks. Generally, mempool policy is more strict than consensus rules and may diverge across nodes. When many nodes agree on mempool policy, it reduces the amount of bandwidth necessary to propagate transactions and blocks across the network, and reduces the overall latency of block propagation, because most nodes can reconstruct the full block from compact block announcements.