I'm not sure. My guess is that these devices have a very small memory, and so that's why it has trouble with that, but I won't swear to that because I'm not absolutely sure.
Discussion
I think it is a choice they stuck with.
If they took a normal next-hop routing approach then they would need a routing table. Meshtastic has no concept of a routing table.
I will take a look at uRNS (which is a C++ impl of Reticulum that runs directly on the ESP)
They have somewhat of a routing table for next hop routing with direct messages, but only for the hop that your node heard the message from. That way, when you send a message in reply, it goes back to the same node that gave you the message to begin with. And then that node passes it off to the node that it heard that message from, etc, etc. But that's only for direct messages and was introduced in 2.6 back in like February 2025. They had a blog post about it if you're interested.
Ah so in certain cases they do, but that is on-demand next-hop routing then?
Reticulum sends announcements at certain intervals and maintains routing table entries via that mechanism.
I would agree with that assessment. It's only four direct messages and it's not maintained for any other purpose besides that.