Your Lightning Simulation Is Lying to You
If you want to test a routing change, a fee policy, or an attack, you can't do it on mainnet, so you do it on a synthetic graph. The problem is that the standard generators don't produce graphs that look like the Lightning Network. Last year I published a comparison of Erdős–Rényi, Barabási–Albert, Watts–Strogatz, and a disassortative BA variant against a mainnet snapshot (11.5k nodes, 41.5k channels). Every one of them misses at least one property that matters for routing: hubs, negative degree mixing, or clustering. I'll go through where each model breaks and why, then show new, not-yet-published results on a simple model that gets all of them right. I'll finish with a short demo of Maestro, the Rust testbed I built to run these experiments: real CLN and LND nodes on whatever topology you feed it, with seeded, replayable payment traffic. Looking for collaborators.
