Having thought about this further I think it’s useful to decouple the problems we’re trying to address.
Follower-based VP
Follower-based VP is a separate concern, and there are all sorts of solutions to this which sit outside of WaterNeuron itself. More and better quality governance participants for followers to follow is the solution with regard to any potential for centralisation in this respect. Better visibility of how much follower-based VP known neurons already have is another way to significantly improve matters here. There’s another thread discussing that.
Direct VP
Direct VP (based on stake) is what WaterNeuron can actually orchestrate and manage (if followers want to follow WaterNeuron’s prevailing vote that’s their choice, they just need more other good options if this is seen as an issue at this stage).
Known Neuron
WaterNeuron’s known neuron is the 8yr neuron, which actually have very little direct VP (roughly 0.1% I think). It’s also not going to change much. I think it’s sensible to leave the known neuron out of this, and allow it to represent a beacon for followers who wish to follow WaterNeuron’s prevailing vote. This significantly simplifies things without really weakening the proposed solution.
6mo Neuron (the guts of the protocol, where there’s room for improvement)
The 6mo neuron is where WaterNeuron’s VP will keep growing, and this is the neuron that could represent a danger to the NNS in the future (if you’re to assume a bad actor gains 51% control of the WTN DAO).
I think as soon as the 6mo neuron’s stake is roughly equivalent to 0.1% of NNS VP, is would be desirable to consider this neuron ‘cold’ and then create a new ‘hot’ 6mo neuron. Once this neuron reaches 0.1%, the same action is taken again (now 2 cold neurons, and 1 hot neuron), and so on. This allows the VP to become more and more granular as the overall VP grows. Not only does this allow direct VP to be cast in a more granular manner that better represents the WTN community on the NNS, it also allows for a nice age bonus optimisation that I proposed back in January (basically there are a whole host of benefits to doing what @infu has suggested).
Age Bonus Optimisation
Copied from Telegram comments I posted earlier this year
The idea would be to allow the protocol to get much much closer to the max age bonus, as well as growing the age bonus more quickly. The idea starts with the observation that if I stake 10 ICP now (to mint nICP) I just contributed an ICP stake with a 0 age (bringing down the average for the whole neuron). This entitles me to unstake that ICP, but when I do that I’ve taken ICP that actually has a much higher age bonus.
If instead the protocol were able to preferentially dip into staked ICP with a lower age bonus (much closer to the age I had contributed), the bulk of ICP in the protocol can just keep aging with minimal disruption
The idea is basically that rather than having one 6mo neuron, the protocol could have several that are managed using a simple algorithm.
There would always be one hot neuron, and zero or more cold neurons (terminology derived from hot and cold storage when archiving data)
In the background, the oldest cold neurons are unlikely to need touching. Once the oldest one reaches 4 years old (and max age bonus), the next oldest will start catching up with it.
[Basically,] If you segregate your 6mo neuron stakes by age (keeping the oldest together and the youngest together), when someone unstakes you can service their request using the younger ICP
By doing that you actually increase the average age for all 6mo neurons (because the amount of younger ICP has shrunk relative to the amount of older ICP)
At the moment the distinction isn’t drawn, so there’s nothing to negate the age bonus erosion that occurs when you stake. Unstaking doesn’t currently increase the age bonus, but there’s no reason that it can’t
Unstaking can partially negate the age bonus erosion of staking if we design it that way (implementation-wise)
Most importantly there would be no change to how users perceive or interact with the protocol, other than APY that tends to be better than the NNS (6mo-neuron-wise), even in the long term!
Related comments about the security benefits of doing this sort of thing (the very good point that @infu is making).