Governed intelligence

Capability is not the same as authority.

WITNESS uses AI inside explicit epistemic boundaries. A model may summarize, compare, infer, and challenge—but it cannot silently promote its own output into established truth.

Governance rules

Unknown is valid

Insufficient evidence can end an investigation without forcing a confident answer.

Model output is not truth

Generated analysis remains distinguishable from source evidence and governed evaluation.

Provenance persists

Evidence retains source identity and relevant collection or provider constraints.

Alternatives remain visible

Competing explanations are preserved when the evidence has not eliminated them.

Uncertainty changes strength

Weak, missing, provisional, or contradictory evidence constrains the conclusion.

Transport is not authority

MCP and A2A can move context and claims without becoming the layer that decides truth.

Governance protects usefulness.

The goal is not to make AI timid. It is to make its confidence proportional to the evidence. WITNESS can still produce strong conclusions when the evidence warrants them, but the reasoning path remains inspectable and revisable.

Governance also protects the creator.

Creator data is not treated as a permissionless substrate. Authorized evidence, principal-scoped access, public/private commitment boundaries, and source provenance constrain what the system may observe, infer, retain, or expose.

WITNESS is not anti-AI. It is AI made answerable to evidence.