V2 P0: Evidence Graph — authority, SUPPORTS/CONTRADICTS, explain API

[cursor]

Background

Signature V2 capability: memory as claims supported by evidence, not opaque facts.

Problem Statement

Confidence is not explainable; conflicting evidence is not first-class; agent inference can look like verified fact.

Proposed Solution

• Configurable source-authority policy (related: Trust scorer source priors configurable per workspace #42 trust priors)
• Write path attaches Evidence to Claims
• Explain API: supporting + conflicting evidence + authority
• Assertion types: ASSERTED / OBSERVED / INFERRED / DERIVED

Acceptance Criteria

• Important memories expose evidence
• Conflicting evidence visible
• INFERRED never silently equals verified fact
• Unit tests: multi-source support, conflict, inference labeling

Depends on

Data foundations (Claim/Evidence migrations)

Plan ref

todo evidence-graph — V2 Enhancement 2