After more than 2 weeks, let me summarize the discussion and propose some edits to the original design based on the received feedback.
We propose the following changes to the original design.
- A few different ideas were shared with respect to what is the right period after which the voting power adjustment is applied. After the discussion, we propose that the voting power adjustment starts after 6 months and decreases until the minimum is reached at 7 months. As pointed out by community members, this nicely means that one relevant action every half year is sufficient for neurons to keep getting rewards.
- For inactive neurons, it was discussed whether in addition to the voting power adjustment the following settings should also be completely removed. Based on the discussion here and because this is closer to the original proposal, we propose to add such a reset of followees to the design.
Let us also address some other alternatives that were raised.
- Decisions could be based on the full voting power rather than the adjusted voting power.
In particular it was discussed whether the current design comes with a significant risk for 51% attacks.- Using the non-adjusted voting power would be a risk for urgent proposals as explained in the “Alternatives considered” of the original post.
- As argued in the discussion, with the new design it is still expensive to launch a 51% attack: sleeper neurons’ tokens are still locked and thus an attacker would have to buy a large portion of the liquid ICP tokens (see Bjoern’s analysis).
- For all actively voting neurons, the relative voting power for decision might increase. For neurons with a lot of (sleeper) followers, e.g., DFINITY, the induced voting power including the followers would decrease as the sleeper neurons would not vote anymore. Therefore, this is overall a step in the right direction.
- The rewards could be based on the adjusted voting power rather than the potential voting power.
- The main reasons why we propose to base the reward on the potential voting power are
- Already today, the voting reward computation is taking into account the non-voting neuron’s voting power. The initial intention of the design was to keep this as similar as possible.
- It can be considered a nice side effect of this feature that it would potentially decrease inflation, as mentioned for example in this post and this message in another post. Likely, as the sleeper neurons would not get any rewards, the effect of this feature is that there is less maturity distributed. If in contrast the adjusted voting power is considered for the rewards, this would increase the maturity, and thus likely increase the inflation, compared to today.
- The main reasons why we propose to base the reward on the potential voting power are
- Rather than having periodic confirmation, there could be a one-off reset, possibly also with batches of neurons being reset at a time.
- The main concern is that it is hard to predict whether it would be possible to make fast decisions in cases of hotfixes. If neurons are reset in batches, it is also unclear what should be done if the first batch is reset and we then observe that it is impossible to reach 50% - would the next batch still also be reset?
- Another motivation for a regular reset that was raised is that users might die or lose their keys. It is thus advantageous to regularly consider who are the voters who are still around and active.
Overall, we think the proposed design with these adjustments reflects the main intent of the original proposal, namely neurons having to regularly reconsider following and a reset of following if this is not done. In addition, it ensures that default following is reset and has the side effect of reducing inflation, which some community members find advantageous.
It is impossible to know what other intentions voters had when voting on the original proposal.
If more details are required to find an agreement, maybe a new motion proposal is needed that replaces the original one.