New DAO Structure Idea - Layered DAO

#1

What is it?

A new type of DAO decision-structure based on a three layer system:

Layer 1: Anyone can submit short ideas. Reputation-holders vote on them, giving each a total budget proportional to its votes.

Layer 2: Anyone proposes execution plans for the layer 1 ideas. Anyone stakes tokens predicting how pleased Reputation-holders will be with the execution of the plan. Plans that are most highly predicted are funded and executed.

Layer 3: Executed solutions are reviewed by reputation-holders, and predictors are rewarded according to the accuracy of their predictions.

Why?

A DAO that:

  • Plans for the future intelligently and emergently

    • General to specific flow of ideas encourages/forces forethought without centralizing or politicizing
  • Makes it fast and easy for reputation-holders to participate

    • Layer 1 reputation voting is budgeted instead of infinitely reusable, so each rep holder’s influence over the DAO is more in line with their rep holdings, rather than the amount of free time they have. You can max out your participation in Layer 1 in just a few minutes.
  • Compresses information about rep-holder preferences and directs the DAO’s actions toward them

    • Layer 1 uses upvoting to direct attention towards preferred ideas (see reddit)
    • Layer 1 uses roots and branches for compressing preferences into clear categories (see Delfy)
    • Layer 2 execution plans are tied to layer 1 ideas, directing the DAO towards those preferences (the tying is game theoretically secure)
  • Gives expertise (with skin in the game) influence over the specific ways those preferences are fulfilled

    • “Voting” (predicting) on execution plans encourages predictors express their confidence through the amount they stake, which should strongly favor people with expertise (You can contrast this with the current Genesis protocol, which favors expert opinions in boosting proposals, but not actually passing them. This system favors experts in the actual passing of the proposals, but punishes them for passing proposals that don’t actually please reputation holders in execution.)

How would it work?

Layer 1 - General Ideas

  • Ideas can be submitted by anyone
  • Ideas use the Delfy system
    • Short maximum length
    • Voting on an idea also upvotes it in terms of visibility
    • ideas have a root and branch system
    • Manages collective attention similarly to reddit (as opposed to the staking/boosting system)
  • Only Reputation-holders can vote on ideas
  • Voting is done by investing reputation into ideas
    • You may only invest up to your current reputation total at once
    • You may remove invested reputation at any time (it isn’t used up permanently; only as long as you keep it invested in an idea)
  • The amount of reputation currently invested in a idea determines the idea’s budget: the proportion of the current DAO holdings that can be spent on it
  • It would also be totally possible to do a budget box style thing here

Layer 2 - Execution Plans

  • Anyone may submit an execution plan
  • Plans can be any length
  • Plans must submitted with budget, a time frame, and a description
  • Execution plans must be tied to a Layer 1 idea
  • Anyone may vote on exeuction plans
  • Voting is done with a prediction game (or some variation of one)
  • The target of the predictions is the execution plan’s follow-up score (the protocol may provide a middle and high expected score to give predictors context)
  • Each layer 2 plan has a prediction score, something like score = sum of each prediction’s predicted follow-up score * amount staked
  • Execution plan passing loop:
    • The plan with the highest prediction score among plans that have
      • been submitted for a minimum time period (this should be related to the collective attention supply of layer 2 participants, something you can at least guess at)
      • a minimum positively staked on them (a constant? something proportional to the plan’s proposed budget so that stakers can’t cheaply pass expensive plans?)
      • a budget within their layer 1 idea’s allotted amount
    • gets “passed” (funded and removed from the list). Rep voters on the plan’s root idea in layer 1 get back some of the reputation they invested (the same as the fraction of the idea’s budget used up by the execution plan).

Layer 3 - Follow-up Rewards

  • For every passed plan, once its time frame has ended, a follow-up period is triggered
  • During its follow-up period, the plan is scored by reputation-holders with a percentage or rating representing its fulfillment of their reputation investment in its layer 1 idea
    • (ie. “From 1 to 10, how much has this plan fulfilled your expectations for [layer 1 idea]?” Or perhaps to present it more directly: “How good of a job did the predictors do choosing this plan to be funded from 1-10?”)
  • Reputation-holders who voted for a idea in layer 1 are rewarded for (or punished for not) rating its descendent passed layer 2 plans during follow-up.
  • An execution plan’s follow-up score is a weighted sum of the reputation-holders’ follow-up ratings (this means predictors are essentially predicting how many people will rate AND what their ratings will be)
  • At the follow up period’s close, predictors on an execution plan are rewarded according to how accurate their prediction was in relation to the other predictions and to the actual score
    • individual reward = ( stake/(1+ f (distance from final score)) for this guess ) / ( sum of (stake/(1+ f (distance from final score))) for all guesses ) * reward pool, where f = some function that affects the slope, so that rewards are spread properly between people who made useful predictions
    • reward pool = sum of all prediction stake amounts + small percentage of the budget allocated to the execution plan (or some other bonus proportional to the budget size — or maybe proportional to the amount staked?)

How does this DAO change its own structure?

Note that most malicious plans would have trouble passing and / or have such little budget that wouldn’t be able to accomplish much.

Let’s consider, though, how the DAO would make changes to it’s own structure, such as adding new reputation-holders or changing the governance schemes. If executed plans have permission to make such changes, the DAO is very vulnerable to malicious structural-change plans. For example, if I make a layer 1 idea like “change the DAO into a monarchy,” invest my own rep into it, and no one else puts rep into it, then that idea will have a relatively small budget. I can still go in to layer 2, however, and create an execution plan for it that changes the DAO’s structure, give it some minimum upstake, and get it executed, as long as my layer 2 plan is the most upstaked plan — there is little incentive for other people to come in and stake down my plan since the budget is very small. It could easily go unnoticed until it’s too late.

So, the initial answer to this question is probably to not give execution plans the power to directly make structural changes of most kinds. You’d need some parallel system that allows the DAO to change it’s own structure. Perhaps structural plans that appear in step 2 go into a Genesis DAO-like structure (predict whether the DAO will pass it, top plans get boosted, then in step 3 voted on instead of reviewed).

For reputation changes specifically, the DAO could employ an automated scheme. It could reward successful proposers with reputation in proportion to their financial rewards as in Colony or reward successful stakers for predicting top-reviewed plans, for a few examples. Automating significant amounts of reputation flow is a bit risky, though, since it means the “shape” of the DAOs membership can change without any “conscious” decision by the DAO — it’s difficult to predict what this might do to the DAO’s mission and behavior in the long run.

Alternate Structures

This layered general-to-specific DAO structure could be applied in very different ways. For example, imagine this profit-focused version:

Layer 1: Anyone can create a short idea (there is no Rep.) Each idea is a bonded curve that anyone can invest in. Holding tokens in the idea gives you rights to future revenue generated by it.

Layer 2: Anyone can propose specific solutions for Layer 1 ideas (they must be linked to a Layer 1 idea, as before). These ideas are also bonded curves that anyone can invest in. Holding tokens in these ideas, too, gives you rights to future revenue generated by them. Layer 1 ideas (and thus Layer 1 investors) automatically own some tokens in their descendent Layer 2 plans (proportional to something or a flat amount?). Layer 2 plans that generate a goal amount of investment automatically launch DAOs to execute the plan, with anyone who owns tokens in the idea having Reputation in the DAO.

(You might want more then 2 layers between initial ideas and fully detailed plans to increase the intelligence of the system, something like: Layer 1, tweet-length ideas; Layer 2, 2 page proposals; Layer 3, fully detailed business plans/white papers/etc.)

Thoughts?

2 Likes
Resource List - DAO R&D
#2

There is a lot of great stuff here. I like how the layered curved bonds impose any process you might want without limiting the paths through.

I’d love to see a few well known and proven processes experimented with in this sort of system.

2 Likes
#3

This is realy difficult to understand. It appears to me to be:

  1. A way to determine the mandate of a DAO
  2. Adjust Rep based on input at the idea stage and after the completion stage of proposals

A mandate of a DAO can be determined by whoever is launching it. It can then naturally grow and change over time naturally assuming there are communication channels around Rep holders, Gen holders, and Proposal writers.

The Rep adjustment mechanisms I do not agree with. The problem is essentially getting the right ideas in place and rewarding or punishing good/bad behaviour.

Half of that is already done more easily by holographic consensus. The other half can be prevented by providing detailed guideline proposals with milestone based distributions of rewards.

2 Likes
#4

Realized I never replied here, only in telegram. Sorry if it’s confusing! I think the summary at the very top is a pretty good simple rundown, if that helps.

Responses to your points:

  1. Yeah, sort of. In theory, it’s a way of finding the “mandate” that acknowledges that that “mandate” is constantly changing as the priorities of the members change and can stay up to date efficiently. I’d argue that’s a better way of doing a mission statement than just setting it at the start for a number of reasons, and there are also many examples of organizations/countries that have benefited from a changeable “mandate” (mission statement/constitution/etc).
  2. I assume you’re getting this from here:
  • Reputation-holders who voted for a idea in layer 1 are rewarded for (or punished for not) rating its descendent passed layer 2 plans during follow-up.

These rewards (punishments) do not need to be in Reputation – in fact, I don’t think it makes sense for them to be Reputation. This is just meant as an incentive for Rep-holders to review executed ideas, and I’m sure it could be structured in all sorts of ways. You probably need some incentive here, because Rep holders have no other reason to review executed ideas, and accurate reviews are necessary to keep the layer 2 prediction game going.

As I wrote in the original post, the advantages of this version of holographic consensus (and I do think this is just another implementation of the same basic philosophy) are that:

  1. It might be better at planning, since ideas go through a development process (layer 1 to layer 2) instead of being full plans being proposed from nothing.
  2. It requires less attention per reputation-holder, or at least a comparable amount, so it’s at least as scalable.
  3. It should direct proposals toward real Rep-holder preferences quicker, since Rep holder preferences are expressed in a completely defined and authoritative way in layer 1 (in our current version, proposers take educated guesses at what the DAO will like, more or less).
  4. It gives experts decision-making power pro rata to their confidence. Rep-holders are experts in what the DAO wants, not how to execute it. This protocol puts Rep-holders in charge of creating the “problems” but lets anyone confident in their skills specific to that problem propose and predict the specific solutions.
1 Like