Link to the talk: Reputation for Machines (EthCC 5)

Reputation is everywhere

• Existed in web1 and web2 platforms and marketplaces like eBay, Uber, Trustpilot
• Provides signals to users of these platforms on the quality of things
  • Builds trust
  • Drives economic activity

Reputation for machines is not new

• These have existed for a long time.
• Things like DNSWL/BL (DNS Whitelist/Blacklist) really have to do with how machines “behave” or “conduct” themselves in a network.
• This reputation also helps machines understand other machines (e.g. their history, counter-party behavior).
• Algorithmic - more first principles; determined set of behaviors that we want to measure with a fixed view of the world
• User-generated - more dynamic; better suited to more expressive situations with measured human behavior such as marketplaces

“Reputation” and “Machines” are not an obvious pairing but they are present in the crypto ecosystem

• Machines run the crypto ecosystem at an infrastructure level (e.g. nodes/validators proposing and attesting).
• These machines are not unboundedly expressive; they follow a spec.
  • Even though those specifications change, they do remain fixed for a period of time until changes are required.
• They also carry through the agency of their operators:
  • How machines and nodes are tuned
  • The way operators decide to take on more or less risk to affect performance
• As these machines perform a set of actions block by block that affect the network as whole, it then becomes important to at least get a sense of or even measure their reputation over time.

The problem with reputation: what exactly is “performance?”

• Specifically when it comes to validator and node operator performance, there are a multitude of definitions - there is no standard.
• The notions of risk and performance in this network are not purely subjective.
  • There is a spec to be followed so there should be some standards on the measure of reputation.
• The lack of standards signals a market failure and a lack of coordination amongst participants.
• Main question: What could you build if everyone agreed on a definition of reputation?

Rated’s view of machine reputation and its components

In the context of Ethereum validators (although transferable to similar network)

Performance

• From the beacon chain/consensus layer spec, validator rewards and their corresponding categories were used.
• Needs to answer the question of how well these validators are doing their duty over time which is to attest to and build the correct version of the chain.
• Contextualize the performance from an atomic (per validator, per block) level and also compose it towards the level of operators.
• Easiest element to capture given data availability: all on-chain, following a set of specifications.

Externalities

• Needs to answer the question of how validator behavior affects other participants in the network even outside of validators.
• Example: How does a node handle MEV (maximum extractable value) post-merge?
  • Validators will have the ability to assemble blocks and with that comes choices:
    • Will they assemble blocks based on gas fees?
    • Would they rather outsource block building?
    • Would they choose to frontrun DEX transactions?
    • Would they choose to sandwich users?
  • These decisions will have an effect on the application level, positive or negative.
• Capturing the effects that stem from the actions of validators is an important part of measuring reputation.

Risk

• Most of the information regarding risk lies outside the chain and is arguably the most complicated to measure.
• Needs to answer the question of how validator behavior is affecting/will affect the network as a whole given the circumstances and context surrounding this validator and its operator.
• Example questions:
  • How many validators are under this node operator and how do they behave?
  • How distributed are the data centers of this node operator?
  • How many clients are these node operators running?
  • How much market share do these set of validators and node operators have and is there centralization risk?

Reputation for Machines: Elevating validators as an asset class

• Currently, nodes are capital assets - software and hardware that produces set future cash flows.
• Attaching reputation to these nodes allows one to explore their potential as financial assets.
• Example use cases:
  • Pricing Insurance for slashing and downtime risk (Nexus Mutual)
  • Onboarding operators based on performance (Lido)
  • Powering distributed validator technology (SSV and Obol)
  • Exploring financial products and derivatives (UMA)
• What can you build if there is a standard measure of reputation for machines? Turns out a lot and there is still so much more to explore.

--

Useful links:

• Rated website
• Documentation
• API Docs
• Twitter
• Discord