Canon term · Auditability

Agent Decision Record

The structured record of a single consequential agent action, sufficient to reconstruct it: the authority under which it was taken, the agent and model state that took it, the evidence available at the time, the reasoning trace, the action committed, and the oversight conditions evaluated.

  • StatusProposed by QIS (not yet industry usage)
  • AdoptionExperimental
  • LayerGovernance
  • Short formADR (Agent Decision Record)

Why it matters

Reconstruction, not logging. Composes with what already exists rather than replacing it — the trace is OpenTelemetry, the order reference is FIX (Financial Information eXchange), the data lineage is OpenLineage. What is missing is the semantic layer binding them, which is the whole proposal.

This is a proposal, not established usage

This term is a QIS Canonical Extension. It does not currently appear in the Model Context Protocol specification, the A2A (Agent2Agent) specification, the AIEOG Shared AI Lexicon, or any instrument in our standards register. We are proposing it because we have documented the gap it fills, and we label it as ours everywhere it appears. Adopt it, argue with it, or ignore it — but do not cite it as industry consensus, because it is not.

Where it comes from

QIS Canonical Extension. Proposed to fill the gap between an execution record, which FIX already carries, and a decision record, which nothing carries.

Proposed here; instruments its origin note discusses

What it is not

These are routinely confused with Agent Decision Record. The distinctions have consequences for how a system is governed.

  • Audit Trail

    A trail establishes sequence. A decision record establishes sufficiency — that everything needed to re-derive the decision was captured.

  • Observability trace

    A trace records the call chain. It does not record model version, memory state, or the authority in force.

How it connects

What it needs

MARQUE questions it addresses

  • AuditabilityCan a decision be reconstructed months later, exactly?

Connected through the MARQUE (a reference framework organized around eight questions) crosswalk, an editorial mapping. These objectives from the Financial Services AI Risk Management Framework are assigned to the same questions that this term addresses. The ones sharing most of this term's vocabulary are shown first, as a reading aid. The assignment is to the question, not to the term: the source does not map this term to any objective directly.

Question: Auditability · 131 control objectives assigned

  • MS-2.5.1Validation and Reliability Testing ProcessMeasure
  • GV-1.4.3AI Documentation Repository and AccessGovern
  • GV-2.1.1AI Risk Management Roles and ResponsibilitiesGovern
All 131 objectives assigned to Auditability

Cite and reuse

Cite the identifier QIS-TERM-00021, not the URL: identifiers are stable, URLs may change. This entry is free to read, quote and index under the dual license. Corrections sent to editor@qisfund.com are published.

Technical details: identifier, graph relationships and record
QIS-TERM-00021: outbound relationships as published in api/graph.json and api/lexicon.json
PredicateTarget identifierTargetPredicate meaning
addressesmarque:auditabilityAuditabilityThe subject speaks to the object as a question or concern.
consumesterm:audit-trailAudit TrailThe subject takes the object as an input.
requiresterm:authority-grantAuthority GrantThe subject imposes the object as a precondition.
dependsOnstd:opentelemetryOPENTELEMETRYThe subject cannot function correctly without the object.
relatedToterm:auditabilityAuditabilityAn association too weak or too general for a stronger verb.
Canonical identifier
QIS-TERM-00021
Tier
QIS Canonical Extension
Domain · layer
Auditability · Governance
Semantic aliases
None recorded
First published
2026-08-02
Last reviewed
2026-08-02 · 180-day cycle