Most reputation systems try to compress trust into one universal number.
This feels intuitive.
But it breaks down immediately when you look at how trust actually works in the real world.
Trust Is Contextual
Think about the people you trust.
You might trust someone to:
-
write smart contracts
-
manage a DAO treasury
-
moderate a community
But that doesn’t mean you trust them equally in every role.
Example:
A trader with deep liquidity experience might be highly trusted for market execution, but that same person may have no reputation as a developer.
A respected open-source engineer might be extremely trusted for protocol development, yet have zero credibility in DeFi trading decisions.
Trust isn’t a single dimension.
It’s domain specific.
Reputation Systems Often Ignore This
Most reputation models collapse trust into a single score.
Examples include:
-
seller ratings in marketplaces
-
follower counts on social platforms
-
aggregate trust scores in identity systems
But these approaches lose important context.
A 95% seller rating doesn’t tell you whether the seller is good at:
-
electronics
-
collectibles
-
luxury items
Similarly, a global blockchain reputation score tells us very little about why someone is trusted.
A Graph Model Allows Context
The Intuition graph introduces a different structure.
Instead of scores, it stores attestations.
Example:
Alice → Bob
Predicate: is skilled in Solidity
Confidence: 0.9
Charlie → Bob
Predicate: is reliable trader
Confidence: 0.8
Now Bob has two separate reputational signals, each tied to a specific domain.
This means reputation can be computed per context, rather than globally.
Contextual Trust Scoring
Once attestations exist in a graph, trust can be computed within a specific scope.
Example queries an agent might run:
-
“Who are the most trusted Solidity developers?”
-
“Which wallets are trusted traders by reputable market participants?”
-
“Which governance delegates are trusted by long-term DAO contributors?”
Each query uses different predicates and different trust sources.
The resulting reputation score changes depending on the context.
Research Supporting Contextual Trust
This concept aligns with findings from reputation system research.
The original EigenTrust algorithm already hinted at this issue.
The EigenTrust Algorithm for Reputation Management in P2P Networks showed that trust propagation works best when it reflects specific interaction domains rather than generic reputation.
More recent work in multi-dimensional trust models also emphasizes this.
Instead of a single trust value, reputation should be modeled as vectors across domains.
This allows systems to capture expertise-specific credibility.
Why This Matters for AI Agents
AI agents increasingly rely on external signals to make decisions.
For example:
-
selecting counterparties
-
choosing developers to collaborate with
-
identifying trusted data sources
But a global trust score doesn’t help much.
Agents need contextual signals.
Example:
An agent searching for developers should prioritize attestations related to engineering, not trading or social influence.
Without contextual filtering, reputation becomes noisy and unreliable.
Possible Design Pattern for Intuition Builders
A contextual reputation system could work like this:
-
Define a domain (e.g., “Solidity development”)
-
Select predicates relevant to that domain
-
Assign weights to those predicates
-
Run trust propagation across the filtered graph
The output becomes a domain-specific trust score.
This approach allows multiple trust layers to exist simultaneously:
-
developer reputation
-
trader reputation
-
governance reputation
-
research reputation
Example Builder Opportunities
Some potential tools that could emerge from this model:
Domain-Specific Reputation Dashboards
Interfaces showing top trusted participants within a domain.
Example:
“Top trusted Solidity auditors according to verified developers.”
Context-Aware Agent Queries
Agents requesting reputation within specific scopes.
Example query:
“Find trusted liquidity providers endorsed by other experienced traders.”
Reputation Marketplaces
Protocols where users can delegate trust or stake reputation within domains.
Example:
Specialized attestation registries for security researchers or DAO governance participants.
Open Questions
If trust becomes contextual, several design questions emerge:
-
How should domains be defined?
-
Should predicates be standardized or evolve organically?
-
Can agents dynamically determine the relevant predicates for a task?
-
How do we prevent reputation fragmentation across too many domains?
A Suggestion for the Intuition Ecosystem
One possible direction is developing domain templates for trust scoring.
These templates would define:
-
relevant predicates
-
weighting schemes
-
trust propagation rules
Builders could then reuse these templates when designing reputation-based applications.
My Final Thought
If the Intuition graph becomes a universal attestation layer, the real power might not be who is trusted globally, but who is trusted for what.
Reputation stops being a number.
It becomes a queryable network of expertise.