one of the reasons I'm hesitant to publish NDK broadly is because right now tooling sucks; all clients are reinventing the wheel

most implementations are naive af (I know, I wrote like five super naive already)

but out of this naivete comes evolution (with NDK I arrived at a very similar conclusion as the gossip implementation where relays are scored and traversed based on prior knowledge (storing this data is similar to the issue LDK has with storing channel state)

I fear that if we release a good-enough implementation that is not stupid we won't feel the pain and continue to evolve to find better relay navigation