01 / capability
Decision trace capture
Exposed as a scoped capability with its own evidence trail, not a hidden setting.
Knowledge & context · Concept
Make the knowledge answerable.
A decision observability system that records how human and AI decisions were formed, authorized, executed and evaluated.
01
What it is
DecisionTracing is built for AI governance teams, compliance officers, platform engineers, enterprise operators.
The problem
DecisionTracing replaces disconnected logs, incomplete audit trails and retrospective manual decision reconstruction.
The approach
A decision observability system that records how human and AI decisions were formed, authorized, executed and evaluated.
The boundary
Work stays inside an explicit scope: named sources, defined permissions, bounded execution, retained evidence.
02
Capabilities
Five contracts the product is specified around. Each one is a boundary as much as a feature.
02 / capability
Causal and evidentiary lineage
Available through the product surface and the machine interfaces, with permission attached.
03 / capability
Tool and action correlation
Defined as an explicit contract so operators can review it before it runs.
04 / capability
Trace validation and replay
Available through the product surface and the machine interfaces, with permission attached.
05 / capability
Observability APIs, dashboards and MCP resources
Exposed as a scoped capability with its own evidence trail, not a hidden setting.
03
How it works
The same three steps apply whether a person or an agent is driving.
01
Define the scope
Start from the boundary DecisionTracing is allowed to act inside: the workspace, the sources, the budget, the authority.
02
Connect the surfaces
Expose the same capability through the interface a human uses and the interface an agent calls, with one permission model.
03
Act with evidence
Every action leaves a record that can be inspected, replayed and challenged before the next one runs.
Read the specification before you believe the category.
DecisionTracing is documented as a concept: requirements, interfaces and evidence expectations. It is not presented as a live service.