This is a critical issue, to cause a vandalism attack, nodes need to stay OK and behave then simultaneously all do the wrong thing, as they do the group may stall a message as they expel the bad nodes. The time taken and costs to perform such an attack do increase over time. So this is a good thing. These attacks cannot damage data but can cause a blip, it’s a lot of effort for such, but folks will try it for sure. So it will be interesting to see it.
We have had a few doctorate students also doing simulation type work like this and there is a very interesting effect. When you run a computational simulation these figures rapidly alter when switched from linear to XOR addressing. It catches everyone by surprise. So there is a very noticeable difference.
As you see if nodes misbehave they get removed, but even getting close to be able to disrupt a group is very difficult. i.e. If a group does not do as it should, all the reporting node needs do is send the same message to a wider group and immediately it can collapse inwards to kill the bad nodes. We have not implemented that logic though, but for sure it’s a facet the network can achieve easily.
You can think of it like this, send an action to group (A) then if result is not in consensus (routing knows) then the same message can go to each member of A but sent to the group of that name. This allows the network to increase a group to X groups where X is number of nodes in a group (32). This provides the X groups with the ability to kick out each nodes that is in error. To aid this each node that sends a message back signs it, this signed message is crypto proof of error and that is enough to deliver to the nodes Manager group to disconnect them.
It does get a bit more complex, but this is the way the network works in XOR space. It allows nodes to be quickly identified and removed. Hope that makes sense, I will write more soon.