Evidence

Witnessed claims and verification

How contributions become evidence — and why evidence is not authority.

Verify-before-sign IMPLEMENTED

An agent submits an unsigned claim through the node's socket API. The node classifies the claim as STATE (mechanically verifiable) or JUDGMENT (human-decidable), then acts:

The node is a verifier, not an oracle — it never signs a claim it cannot confirm, and it never invents a result.

src/claims.rs

Agentsubmits an unsigned claim
Nodeclassifies claim type
STATEmechanically verifiable
Verifierruns a registered check (e.g. build-result)
true → sign + stake
false / uncheckable → refuse
JUDGMENThuman-decidable
Nodesigns without stake
Effectsigned evidence feeds thickness / contribution
Verify-before-sign claim flow. The node is a verifier, not an oracle — it never signs a claim it cannot confirm. Source: src/claims.rs.

Witness panels IMPLEMENTED

Contributions are attested by a witness panel selected through deterministic, thickness-weighted sortition. Panel selection is a pure function of the committed prefix, so honest nodes derive the same panel. A density guard rejects panels drawn from too sparse a mesh.

src/sortition.rs

Claim taxonomy IMPLEMENTED

The STATE/JUDGMENT boundary is codified in a ratified claim taxonomy. STATE claims carry typed fields (never prose); an unknown claim type routes to a human by closed-world default — the system fails safe toward more human attention, not less.

docs/taxonomy-v2.md

Evidence is not authority

Witnessing creates evidence, not authority. A witness count proves that work was observed and verified; it does not by itself confer standing or the power to ratify. Influence (thickness) is derived from contribution; authority is a separate, still-under-construction concern (see Development).