owned this note
owned this note
Published
Linked with GitHub
# Analytics & Research | 24 May 2022 | 15:30 UTC
###### tags: `analytics` `research` `meta-gov`
# **Agenda**
### Introduction to Analytics & Research subWG
- **Analytics**
**1.** Responsible for the gathering of intel and analytics of all things pertinent to ENS DAO (not ENS Protocol; relevant subWG for that should exist in Ecosystem WG). Currently, *small* fraction of meaningful metrics are already provided by [ENS Tools](https://ens.tools), ENS Leaderboard, ENS Sales Bot, [ENS Dashboard](https://datastudio.google.com/u/0/reporting/8785928a-71d5-4b17-9fea-fe1c937b064f), [ENS Dune Dashboard](https://dune.com/makoto/ens) that are pertinent to Governance.
**2.** [Flipside Crypto](https://flipsidecrypto.xyz) is arguably more equipped to provide relevant data for ENS DAO via their open CEA workstream. All analytics will be performed in-house; third party providers shall only be responsible for providing raw data.
**3.** [ShowKarma](https://showkarma.xyz) is relevant from a community engagement perspective since they will add to user, delegate and contributor engagement.
**4.** Besides the already existing data hooks mentioned above, A&R must also actively seek to pull, archive & analyse data pertaining to off-chain and on-chain votes, ENS token dispersion, Discourse data, Discord data and Twitter analytics. The aim of this exercise is to equip the DAO with all the necessary data and information required in its several decision-making processes.
- **Research**
**1.** A&R is responsible for briefing all working groups with data-driven approaches to streamline ENS DAO. One example of a duty of A&R toward ENS DAO is to formulate a dynamic and machine-generated model for capturing a reputation model for DAO delegates, stewards and contributors. Currently, most reputation models are static, manual, rudimentary and gameable. An ideal reputation should be not gameable or static. In this context, A&R should look into already existing approaches such as classic *Hedonic regression* and Hedonic-AI regression to rid us of manual and gameable models; this is possible but needs dedicated work to formulate training datasets (for Hedonic-AI) and quantifiable data for fitting (classic Hedonic). This work requires devising detailed semi-annual surveys to capture the state of ENS at the very least in consultation with the ENS DAO community and contributors. The result of this undertaking should be to provide the DAO with actionable intel during elections (e.g. reputation of candidates) and other relevant decision-making processes.
**2.** A&R is also responsible for actively assessing threats to ENS and propose ideas to make the DAO resistant to risk and volatility; an example of this is to devise practical ways of reducing dead votecount, off-boarding delegates & contributors, and suggesting improvements to governance model.
### Roadmap
#### Q1/2:
- Steward & Delegate Health Cards
- ENS DAO Governance Report for Q1/2 2022
#### Q3/4:
- Ungameable Robust Reputation/Karma
#### Quick introduction to Ungameable Reputation
In typical gameable karma models, the *karma factor* is assumed to be
```
gameable_karma = (W1 × X1) + (W2 × X2) + (W3 × X3) + ...
```
where `X = [X1 , X2 , X3 , ... ]` are *assumed* properties that affect karma, e.g. proposals initiated, proposals accepted, like count, daily engagement etc (limited engagement data), while `W = [W1 , W2 , W3 , ... ]` are relative weights of those properties. `W` is usually *assumed* using "intuition".
#### How can one design ungameable models?
In contrast, ungameable models should look like
```
robust_karma = (W1 × X1^E1) • (W2 × X2^E2) • (W3 × X3^E3) + ...
```
where `X` are derived from surveys and *not* assumed from available metrics, `•` are operations (*not* necessarily `+`), and `W` & `E` are dynamically calculated by fitting the survey data to the model and *not* assumed (e.g. using non-linear regression). In this context, the surveys must be designed to produce the necessary quantifiable data as well as the variables featured in a robust karma model.
**tl;dr** Ungameable and Robust models are derived from surveys with minimal assumptions, and are *dynamic* due to being community driven, which makes them ungameable. This also means that the quality of ungameable models depend entirely on the quality of surveys.
### TO DO NOW!
#### What's the plan for the upcoming elections? (before June 5)
- Stick to the basic gameable model for this term. It is restrictive but it should be relatively sane since a model is only gameable after it has been defined; therefore the first iteration is not gamed. Engagement data from ShowKarma.xyz [@mmurthy]
- Input from serial call attenders [@BrianMillsJr]
- Offset the gameable bias with a DAO-wide survey sometime during June 1--5 and publish those results in the Steward and Delegate Health Cards. [@RnDAO?]
#### Post-elections and before Q2/3? (June 15-30)
- ENS DAO Governance Report for Q1/2 2022 to be published by the end of June. Flipside Crypto makes significant contribution to the report [@fig]