Simulate, Publish, Activate & Trace
Simulate the Operator DAG without infrastructure mutation, publish versioned Operator revisions, activate one for STRATUM AI, and trace how a Mission moved through policy, approval, execution and verification.
Design → Simulate → Publish → Activate → Trace
AI Studio separates design-time editing from runtime use. Build or change the Operator in DESIGN, test its reasoning path with SIMULATE, publish a versioned revision, activate the revision STRATUM AI should use, and open TRACE when you need to understand what happened during a Mission.
Published revisions give the Operator an identity that can be associated with the Missions it produced. That makes “why did the AI do that?” an answerable operational question instead of a conversation-memory problem.
The downstream DAG
Execution & Output
- AI Planner — Consumes intent plus specialist recommendations, policy and authority inputs and produces the proposed plan.
- Policy Gate — Evaluates the proposed plan against policy and authority before it can continue toward approval/execution.
- Approval — Represents the human-approval stage for mutations that require an operator to Apply the plan.
- Mission DAG — Compiles the approved plan into the runtime Mission dependency graph: ordered and parallelizable execution steps with explicit dependencies.
- Verification — Checks the result using the evidence and checks required by the Operator so success means the outcome was proven, not merely that commands returned.
- Reconciliation — Compares actual state with Desired State after execution and provides the correction/reconciliation stage when the system drifts or verification exposes a mismatch.
- Desired State — The output of the Operator: the target infrastructure/application state that execution and reconciliation are trying to make true.
Test the Operator without changing infrastructure
SIMULATE accepts a test request and evaluates the Operator DAG without performing infrastructure mutations. Use it before activation to see which Roles, Skills, guardrails and policy stages participate and whether the proposed path is allowed.
During simulation, AI Studio animates the Operator DAG so the dataflow is visible. That turns policy and specialist behavior into something an administrator can inspect before handing the Operator a real Mission.
Simulation is for evaluating the operator graph and policy path. It does not perform the infrastructure mutation being described by the test request.
Make the Operator a controlled release artifact
Save Draft while the DAG is still being edited. Publish when the graph is ready to become a versioned Operator revision. Activate the revision that normal STRATUM AI requests should use.
The revision model gives AI behavior the same operational discipline as the rest of STRATUM: changes are explicit, publishable and traceable rather than silently changing because someone edited a hidden prompt.
Explain the Mission through the DAG
TRACE uses the same visual DAG as a read-only explanation surface. Active nodes and edges show which specialist, policy and execution stages affected the Mission. Click into a node to review why it participated, what it contributed and what evidence or policy result came out of that stage.
This is where the DAG becomes more than appearance: the connections represent real runtime semantics, and the same graph can be used to understand why policy or specialist behavior affected a Mission.
From operator reasoning to deterministic work
The Operator DAG is reusable governance and intelligence. The Mission DAG is the per-request execution plan produced after planning, policy and approval. Its dependencies let STRATUM order work safely, run independent steps in parallel, pause/resume/cancel execution, and verify the final outcome against Desired State.
The result is a complete chain: User intent → Operator DAG → Plan → Policy Gate → Human Approval → Mission DAG → Verification → Desired State / Reconciliation.