Trust Isn't a Score. It's a Graph

3. Reputation computation
1

Most systems treat trust like a number.
This person is 0.82 trusted.

Clean. Simple. And lowkey misleading.

Trust doesn’t exist in isolation.

You don’t just trust someone. You trust them because of who you trust.

Example:

  • You trust Billy

  • Billy trusts Zet

  • You’ve never interacted with Zet

But now - you partially trust Zet. That’s how trust actually works.

This is exactly what EigenTrust models.

Trust flows through a network. Not directly but through connections.

  • Billy trusts Zet

  • Zet trusts Luda

  • → Billy indirectly trusts Luda

Now it’s not just about Luda alone. It’s about who stands behind Luda.

Not all trust paths are equal.

Billy → Zet (strong) → Luda (strong)

vs

Billy → RandomUser (weak) → Luda (weak)

Both paths reach Luda. But the confidence is completely different.

Trust depends on: strength of relationships, number of hops, quality of the path.

Why this breaks most systems.

Most reputation systems rely on direct ratings, averages, aggregated scores — but ignore network structure.

Two people with the same score can have completely different trust backing.

The graph changes everything.

With Intuition attestation graph, you don’t just read trust — you traverse it.

Queries that are now possible:

  • “Who do trusted devs like Zet trust?”

  • “Which new builders are one hop away from high-trust nodes?”

  • “Is Luda trusted by the right people for this specific task?”

Why this matters for agents.

Instead of: “Luda has a trust score of 0.91”

An agent can reason:

  • Luda is trusted by Zet

  • Zet is trusted by Billy

  • Billy is a high-trust node in the network → Luda is likely credible

That’s not just trust. That’s explainable trust.

What this unlocks for builders:

  1. Trust Path Explorer — show why someone is trusted, not just the score

  2. Second-Order Discovery — find people trusted by people you trust

  3. Trust Circles — clusters of high-signal relationships

  4. Path-Based Scoring — rank trust by who the path goes through and how strong each link is

Combine graph traversal + contextual filtering + weighted predicates and you get programmable trust.

Open questions I’m thinking about:

  • How many hops before trust becomes noise?

  • Should some nodes carry more weight than others?

  • Can agents dynamically choose the best trust path?

  • How do we visualize trust paths clearly for end users?

All of this is live and explorable today at mcp.intuition.box

What kind of trust paths would you want agents to actually use?

2 Likes