1ML shows larger number of nodes comparing to other lightning network explorers

Question

I found that 1ML displays almost double the number of nodes comparing to other lightning network explorers.

I think 1ML does not provide the datasource but are you able to guess about the reason of the difference?

Besides 1ML, the numbers of nodes are almost same.

My own c-lightning nodes shows about 4,000 by listnodes command and lightning.chaintools.io uses lnd's GetNetworkInfo rpc call according to their source code.

• 1ML https://1ml.com/statistics

Nodes: 7,462
Channels: 39,310

• ACINQ https://explorer.acinq.co/

Nodes: 4,011
Channels: 39,514

• Indexplorer https://graph.lndexplorer.com/

Nodes: 3,941
Channels: 38,983

• Lightning Explorer https://lightning.chaintools.io/

Nodes: 3,971
Channels: 39,201

• Lightning Explorer https://explore.casa/

Nodes: 3,978
Channels: 39,009

Answer 1

Let's start by ignoring the 1ml case. It is perfectly reasonable that the number of nodes differs a little bit as the gossip protocol takes some time to forward the information. In particular there could be private nodes having a channel to the node that is used by the explorers and they could include the knowledge about such a node in their counting...

So why is 1 ml counting almost twice as many nodes? As it is closed source we can only guess. I observed that 1ml also depicts closed channels for a node. So I could imagine that the number of nodes shown by 1ml also includes nodes which existed in the past by counting all nodeids they have ever seen in their database instead of counting the nodes from the gossip store. I did not find such an example quickly. Also the gossip protocol specifies to remove channels if they close on chain but has as far as I rember no rule for removing nodes. Maybe 1ml uses a different / secret lighing implementation. That never drops nodes from their gossip store.