Documentation

What is trustgraphs?

Trustgraphs is a way to compute useful results from verifiable, relationship-shaped data without asking everyone to trust the computer that did the work.

A community or application chooses the data, the rules, and the result it needs. Anyone can run those rules. A zero-knowledge proof shows that the published result came from the committed inputs and the agreed program.

A trustgraph is not one particular social graph#

The name describes a pattern, not a universal network or algorithm. A trustgraph has three parts:

  1. Verifiable inputs. The source might be onchain attestations, signed social records, an anchored repository history, or the proven output of another network.
  2. A deterministic program. Public rules turn those inputs into scores, allocations, or another result. The same inputs and rules always produce the same output.
  3. A proof and an output commitment. The proof checks the computation. A compact onchain root lets an application verify an individual result without storing the full output onchain.

The proof establishes that the program ran correctly over the inputs it was designed to authenticate. It does not decide whether the source data is meaningful or whether the community chose good rules. Those remain part of the network's trust model.

The main implementation today: vouching#

The standard Trustgraphs network uses public Ethereum Attestation Service (EAS) attestations. Members create weighted, revocable vouches between Ethereum accounts. Those vouches form a directed graph, and a seeded PageRank-style program turns the graph into reputation scores.

This is the primary path supported by the network creation flow today, but vouching is one way to produce a trustgraph, not its definition. How vouch scoring works explains the behavior and limits of that program.

Other inputs and results#

The same architecture can support programs with different semantics:

  • A contributions round authenticates a parent network's vouch history alongside claims, responses, and peer evaluations. One proof recomputes reputation, uses it to weight the evaluations, and produces a funding allocation. A contribution claim or evaluation is not a vouch.
  • AT Protocol or Nostr programs can authenticate signed records and define their own rules for turning those records into graph edges and scores.
  • A composition can combine several proven score sets without merging their raw source data.
  • A signer-selection program can recompute ranking from checkpointed vouches, combine it with direct-governance activity and current Safe state, and propose a Safe owner set.

These programs share proof infrastructure, but they do not all share the vouching model or the same algorithm. Their source authentication and maturity also differ; each program's documentation defines what its proof actually checks.

Keep reading#