Great summary, but I wish I would've seen this before finishing my initial rough implementation for #electrum last week :) Getting a full picture of the spec requires quite a lot of trawling through PRs and scattered snippets of pseudocode.

The concatenation of route to introduction point and the blinded path took some time to grasp, but the test vectors in the PR are a nice validation target to work towards.

How is CLN finding routes over the network to the target node/ introduction point? In Electrum we currently just use the channels from the channel graph as routing edges, but in theory this is not strictly needed. However you can't expect nodes to make new peer connections just to satisfy a onionmessage forward. There's currently about 445 nodes advertising support for onion messages, so the graph is not yet very traversible, and there may even be multiple disjointed subgraphs (need to check).

Next challenge for #electrum will be sending/receiving onion messages without a channel graph, over trampoline...