Provenance, Lineage, Causality & Evidence

Data lineage explains information history. Execution Lineage explains application behavior history.

Connect source data, retrieved context, calculated facts, identity, authority, policy, decisions, human approvals, durable state, capability attempts, and external effects under the same Execution Instance.

Source to context query, retrieval, features, Context Package
Decision to effect policy, approval, operation, attempt, receipt
Evidence is written during execution, not assembled later from whatever was logged.
Evidence substrate

Preserve evidence in the Data Plane

The AI Data Plane preserves or references evidence artifacts and typed lineage relationships that survive execution.

Explore Behavior Intelligence
Investigation surface

Investigate behavior through Behavior Intelligence

Behavior Intelligence uses those records to inspect one execution, compare many, reconstruct historical behavior, and perform governed replay.

Precise relationships

Keep provenance, lineage, causality, and evidence distinct

Provenance
Data Lineage
Context Provenance
Decision Lineage
Execution Lineage
Causal Binding
Evidence Receipt
Inspection Projection
Stable anchors

Use the Published Graph and Execution Instance as evidence anchors

Evidence becomes easier to join when it attaches to the application version that was permitted to run and the realized execution that actually ran.

Published GraphExecution InstanceContext PackageDecision-Time Feature SnapshotPolicy DecisionHuman Approval StateBusiness Operation IdentityEffect Receipt
Context Provenance

Preserve what crossed the decision boundary

Origin, eligibility, selection, exclusion, transformation, freshness, order, truncation, and final use remain connected to the downstream Decision Object.

Explore Context Assembly
Decision Lineage

Preserve the chain of judgments

Deterministic validation, model proposals, sufficiency gates, policy obligations, routing, human approval, revalidation, and final authorization stay connected.

Execution Lineage

Carry authority, state, versions, and effects into the evidence record

Authority context

Subject, actor, workload, delegation chain, tenant, legal entity, purpose, assurance, policy version, and obligations.

Durable transitions

Published Graph version, proposal revision, wait state, arriving event, revalidation result, and resource revision.

Operation to effect

Invocation proposal, Business Operation Identity, attempts, timeouts, retries, reconciliation, compensation, and Effect Receipt.

Causal bindings

Preserve typed relationships, not only chronology

A timeline shows sequence. An evidence graph shows how one artifact formed, constrained, authorized, attempted, or confirmed another.

derived_fromselected_intoinformedconstrained_byrequired_reviewapprovedattempted_asconfirmed_by

Map the evidence spine for one consequential AI outcome.

Use an architecture review to define source-to-effect lineage, evidence coverage, causal bindings, permission-aware inspection, and integrity requirements.