For sure. In this comment I'll be using "validators" to mean "nodes participating in consensus". The "true" chain according to the new protocol is the one that

For sure. In this comment I'll be using "validators" to mean "nodes participating in consensus". The "true" chain according to the new protocol is the one that has more signatories. Validators sign blocks as they're produced, and when a block gets more than 2/3rds of the signatures, it's finalized on the chain and won't be rewritten. (Well, it's much more complicated than that, but that's the general intuition.) When you sign a block, you get a reward on that chain, to incentivize people to participate in consensus. So the question is, when there's a fork, what stops people from just signing blocks on both forks so they get the signing reward no matter which fork "wins"? The protocol disincentivizes this by requiring validators to put up a stake of 32 eth, that gets "slashed" if you do that. Put another way, if there's a fork and you sign blocks on both sides of the fork, you lose part of your stake. But that threat only works if you actually have a stake to lose. If it's been ${withdrawal_delay} months and you no longer have any eth staked, you can start signing blocks from a long time ago and the protocol can't slash you as punishment. So a group of 2/3rds of the former-validators could freely start a fork from ${withdrawal_delay} months ago, sign a bunch of blocks on it, and if their fork loses they'd face no penalty. Someone currently participating in consensus can see this is happening and won't be confused, but someone just joining the network will be. The solution is pretty easy, if inelegant. When entering the network for the first time, if there are multiple competing forks, you'll just find someone you trust IRL and ask them which chain is the real one. If you've been offline for less than ${withdrawal_delay} months, you won't have to worry about this, but it is a problem for fresh users. If you set ${withdrawal_delay} to infinity, this isn't a problem, but probably you don't want to do that. Once withdraws are enabled, there'll be some tooling to make it easy to figure out which chain is the real one, but I don't expect there'll be many forks that are remotely convincing.