Trust Model
The trust model is intentionally narrow.
Peers
Peers are authoritative only for their own facts.
They may:
- publish roots
- disclose facts with proofs
- verify bundles from other peers
- publish certification facts in their own trees
They may not:
- rewrite another peer's facts
- claim semantic authority over another peer's tree
Verifiers
Verifiers should be able to:
- inspect a received proof bundle
- verify only the referenced slices
- evaluate the shared claim from explicit rules
They should not need:
- hidden operator state
- the full global database
- a privileged coordinator to tell them what is true
Witnesses And Publication Layers
Witnesses, feeds, relays, and settlement layers may help with:
- ordering
- timestamping
- publication
- escrow
They should not:
- own the only meaningful state
- decide semantic truth for peer-owned facts
On-chain root anchoring is the strongest form of publication layer we expect to use. It makes peer root publication public, ordered, and durable, but it does not make peer-authored facts true by itself. See On-Chain Root Anchoring.
Certifiers
Certifiers are still peers.
Their role is special only in that they verify an input bundle and promote the result into a new certified fact. That does not make them a central oracle.
The protocol must say whether downstream users:
- trust the certifier directly
- audit through the certifier to the inputs
- or support both modes