Transparency Capabilities
Status: first pass
Transparency Capabilities are producer-advertised actions or machine-readable affordances that help a Viewer, Operator, Verifier, or agent inspect and refresh trust state. Surface exposes these capabilities; producers perform the underlying domain work.
Capability Types
Minimum capability categories:
- evidence reverification: recheck source state, rerun a test, refresh an observation, or validate an attestation
- authority reverification: confirm that an actor or system still has authority
- integrity check: compare current source state with the integrity reference behind evidence or claims
- evidence access request: ask for permissioned or private evidence
- materiality review: ask an Operator to review whether a claim is material to a product output
- dispute or correction: report a challenge to a claim, evidence item, or source
Capability Record
A capability should describe:
- stable capability ID
- capability type
- producer that owns execution
- applicable claim, evidence, policy, subject, or claim group scope
- availability state
- input requirements
- expected output shape
- audit trace or event emitted after execution
- human label and suggested action text
Surface tools may render or expose the capability through Trust Panel, Console, API, or MCP resources. Surface must not imply that availability guarantees a successful refresh or that Surface performs the check.
Agent Guidance
Agent-readable guidance may recommend actions such as reverify, ask for evidence, ask a human, surface uncertainty, or do not rely yet. Vertical products and agent runtimes own execution policy and final action decisions.
Current Implementation Names
Use current field names only where exact technical reference requires them:
collectWhenremains a verification policy signal that producers may use to schedule evidence collection.evidence.executionremains a current trace shape for command or MCP tool results.- producer-specific capability details may live in
metadatauntil a portable schema is added. - MCP integrations should expose the same trust state rather than replace the Open Trust Format.
Non-Goals
- no guaranteed refresh
- no Surface-owned verification service
- no agent runtime policy engine
- no certification or approval decision