Table of contents

Introduction

My background

• Designer, programmer & systems architect
• 7+ years experience building a large, profitable & toxic community CEO.CA (60,000+ members)
• Self-funding & using learnings from CEO.CA to focus on next gen product, Spiel

Spiel project status

• Working Beta version is public, missing some features - https://beta.spiel.com
• iOS App ready, Android app imminent
• Spiel v1.0 on track to launch soon
• Codebase will be open-sourced once the protocol is mature

Core design principles

• Skin in the game
  • Communities must be governed by those with finger on the pulse
• Fun & freedom
  • Allow people to create culture, not force culture on them
• Evolution as core mechanism
  • Many parallel experiments allow for best governance systems to survive
  • Communities should be easy to "fork" in case of bad governance

Motivations

• Automating moderation with fixed algorithms turned out to be impossible
• Centralized moderation is exhausting and toxic for the community
• Fixed reputation algorithms are biased and vulnerable to abuse - hard to scale
• Moderation needs to be an immune system, not an algorithm

Low-hanging fruits

• Networked thought needs to be taken social
• Lots of opportunity in backlinks, tagging & forking conversations
• Custom search filters as subjective feed algorithms
• Maps as online social spaces (MMOs get this)
• Progressive rich text editing & flexible embeds
  • Combine chat/blog/notes/wiki concepts under one UI

Implementation details

Social shared reality / simulacra

• "What is this place? Where am I? What rules apply here?"
• Concept of "a place" is important
• Nested structure allows for paths (addressing)
• "Where am I?" should have a meaningful answer
  • On Twitter, each post is its own context
    • Recursive but not addressible
  • Slack/Discord/Reddit have flat structures
    • Addressible but not recursive
  • Filesystems & URLs know what's up
    • Recursive & addressible by path
    • ~channels/can/have/infinite/depth

Inehritance within the hierarchy

• Top-level channel sets initial rules
• Public channels inherit rules from parent
• Users can propose & pass local rule changes
• Independent/private channels start new jurisdictions
• Private group chats & DMs can also use nesting within them

Tagging system

• Tagging a #channel cc's your post there (unless opted out)
  • Creates a distribution/discovery layer for conversations
  • "Surfing conversations", not just "catching up"
  • 
• Relative/absolute channel addressing syntax inspired from Bash:
  • ~realm - top-level channel (community root)
  • ~realm/hashtag/nesttag/etc - slashes for infinite depth
  • #hashtag - shortcut for first nesting level
  • >nesttag - nest into child of current channel
  • <parenttag - refer to parent of current channel
  • ^siblingtag - refer to another channel in the same context
  • @atmention - user profiles are their own top-level channels
  • [[wiki links]] - personal tags, always nested under author's profile (@murat/wiki_links)
  • #private.🔒and #independent.🏛 suffixes to prevent clashing
  • Reply threads are sub-channels for individual posts

Profiles as channels

• User Profiles as index for all public posts by user
• Combines Twitter-style "follow" concept with chat architecture
• Subchannels in user profiles act as personal bookmarking, notetaking & publishing
• Allows following only specific topics by a user (if used diligently)

Maps

• Channels optionally placed on map as polygons
• Physical maps as channel discovery layer
• Maps are ideal for nested contexts (zoom/pan)
• Spatial relevance and finite space create fitness functions
• Social Earth map & fictional maps allow for new types of shared contexts

Governance

Democratic governance

• Formalized into simple elements
• Voting and polling as widgets within posts
• Vote analytics for better visibility into collective consciousness
  • Breakdown of poll results by role or other filters is useful even if not directly tied to policy changes
• Users propose policy changes (motions) & vote on them to steer culture
• We can support all possible governance systems under one framework
  • Most things start as dictatorships, then slowly turn into hierarchies
  • The necessary feedback loop is formalized into a policies system:

Policies for channels & jurisdictions

Policies implement policyscript, which is very similar to ElasticSearch's JSON query sytax. It's a Lisp-like DSL that is mostly concerned with boolean outcomes to nested functions. It is a way of leveraging built-in functions to compose highly specific moderation frameworks. This is made user-friendly by providing a Scratch-like drag-and-drop GUI.

Most of the time, users will pick their policies from ready-made templates; however power-users are encouraged to compose new policy templates.

• Reactions on posts. Can be used for search filtering & scoring.
• Democratically curated list of reactions per context
• Reactions being tied to real consequences is key for tangible culture
  • i.e. In a channel that encourages shitposting, 5+ "😐 Too serious" reactions punish the poster by silencing them for 3 hours.
• Roles assigned/revoked as a function of user activity, reactions, etc.
  • i.e. User gets a Bouncer role if Member for 3 months and 10 other members vote for it
  • i.e. User gets Banned role if they've received the Troublemaker role more than 3 times
• Permission policies are boolean expressions that determine what a user can do based on activity, roles, etc.
  • i.e. Users with Bouncer role can ban Members younger than 1 week
  • i.e. Users with Troublemaker role can post at most 1 message per hour
• Governance policies dictate voting thresholds for accepting motions to change all policy types, including itself
  • i.e. Pass if at least 40% members vote "Yes", and zero moderators vote "No"
  • i.e. Reject if 30% vote "No" or 5% vote "Absolutely NOT"
• Funding policies as formulas for distributing income to mods
• i.e. Income is distributed equally among moderators, proportional to number of days online

The Lisp-like syntax leaves it up to users' creativity to combine core functions and compose fine-tuned channel policies.

Moderation toolset

• Scriptable roles & permissions cover all possible moderation use cases, such as:

  • Banning, temporary silencing & throttling
  • Trial periods
  • KYC requirements, or forced anonimity
  • Allowing or disallowing alt handles
  • Penalty boxes (can only post in #X channel for Y hours)
  • Limiting policing (i.e. can delete max 5 posts per day)
  • Activity throttling (i.e. regular members limited to 20 posts per day)
  • etc...

First-principles design that covers all possible use cases

• Support small & large social spaces, personal notetaking, bookmark organization & publishing & wikis all under one UI paradigm

• All these use cases are emergent from the contextual policy system

• Ended up being very little code

Next steps

The spice must flow

• Three types of value creators: Devs, Mods, Posters
• Monetization is for devs, Markets are for mods + posters
• Moderation is incredibly taxing & creates much value, needs compensation
• Per-channel (or jurisdiction) paywalls & fundraisers where necessary
• Users can vote on funding policies to compensate mods
• Planning on using BTC (Lightning Network) primarily

Customization beyond bots

• Custom themes & widgets are essential to communities
• Individual posts or sidebars can incorporate interactive widgets
  • Imagine stock charts, sports standings, live streams, e-commerce
• Allow people to write their own plugins
• Become the Wordpress of chat

Road to decentralization

• All permissions, roles & cross-tagging extend gracefully to federated third party hosting
• Step 1: DNS-based federation (e-mail, IRC, mastodon)
  • Allow/deny lists based on domain name
  • Permissions based on remote roles & reputation
  • Users trust servers (like we do with e-mail)
• Step 2: Actual DAOs
  • Global single sign-on & cryptographic proof of authorship
  • Doesn't matter which client you use
  • Policies enforced collectively (auditability, tamper-proof)

Other topics to consider

• Encryption - Hairy topic, can get into this more later.
  • Potentially doable with minimal effort if acceptable to leave metadata unencrypted
• Supporting existing chat protocols (Matrix, ActivityPub)
• Leveraging AI-assisted rule & policy creation to increase user participation in governance