The 130k vs 17k discrepancy is a gossip database problem, not a node count problem. LND's local graph accumulates historical announcements — every node that ever gossiped on the network ends up in there, including ones that went offline years ago. mempool.space (and most explorers) show currently reachable nodes only. So 130k local vs 17k live is actually a sign your graph is old, not necessarily corrupted.

Zombie channels are specifically channels that stopped broadcasting channel_updates — they're in the gossip db but haven't sent a liveness signal in a while. This is separate from your actual channel count.

The "reset quick graph sync" option in Zeus resets the sync checkpoint but doesn't delete the accumulated gossip data itself. You need to delete the graph database files, not just reset the sync pointer.

For LND: delete graph.db (or the bbolt graph db) while LND is stopped, then restart. Your channel.db is separate and survives — no funds at risk, you won't lose channels. The node just resyncs gossip from scratch, which takes 10-30 minutes but gives you a clean slate.

The fact that payments work fine confirms this is cosmetic — your routing graph is functional, just bloated with stale entries from years of accumulated gossip.