Governed Mutation Pipeline Demo (Finance Example)

An AI system can move $2M without approval — unless something stops it.

This demo shows a simple but critical problem:

An AI proposes a financial change… and executes it in the same workflow.

No separation of roles.
No required approval.
No control at the point of action.

The change just happens.

---

What This Demo Shows

This demo runs the same financial action twice:

❌ Without enforcement

• The AI proposes and approves the change
• Required roles are violated
• The system executes the action anyway
  → Funds are reallocated

---

✅ With CRI-CORE enforcement

• Role separation is enforced
• Required approvals are verified
• The action is evaluated at the execution boundary
  → The change is blocked before it happens

---

What This Is

This is a deterministic execution control layer.

It sits at the exact point where a system attempts to act and decides:

• ✅ allow execution
• ❌ block execution

No warnings.
No after-the-fact auditing.

The action either happens — or it doesn’t.

---

Install

pip install cricore

---

Run

python -m runner.run_demo

---

Scenario: Finance — Budget Reallocation

An AI system proposes reallocating budget between departments.

Required Roles

• proposer → AI system
• responsible → Finance Manager
• accountable → CFO

Rule

Separation of duties:

responsible and accountable must be different people.

---

What Happens

❌ Blocked Case (Invalid)

Finance Manager is assigned to both roles.

independence: FAIL
COMMIT BLOCKED

Result:

• No funds are moved
• The action is stopped at the mutation boundary

---

✅ Allowed Case (Valid)

Roles are properly separated.

independence: PASS
COMMIT ALLOWED

Result:

• Funds are reallocated
• Execution occurs only after authorization

---

Expected Behavior

When you run the demo:

• Without CRI-CORE → both actions execute
• With CRI-CORE → unauthorized action is blocked

Execution is no longer assumed.
It is explicitly authorized or blocked at the mutation boundary.

---

How It Works

Proposed Action
↓
CRI-CORE Enforcement
↓
commit_allowed = true / false
↓
Execution (or no execution)

CRI-CORE acts as a control point, not a validator.

It determines whether execution is permitted.

---

Where This Applies

This pattern can be used anywhere a system performs real actions:

• AI agents and autonomous systems
• financial workflows
• CI/CD pipelines
• compliance-critical operations

---

What This Proves

• Unsafe actions can be stopped before execution
• Governance rules can be enforced deterministically
• Responsibility can be verified structurally
• AI systems can be controlled without removing autonomy

---

Who This Is For

Engineering / Platform Teams

• Enforce constraints at deployment or commit boundaries

AI / ML Teams

• Control AI-generated actions before execution

Compliance / Risk Teams

• Enforce separation of duties and accountability

Organizations deploying AI

• Move from monitoring → enforcement

---

Status

Demonstration of CRI-CORE as an execution-boundary enforcement layer.

---

_{© 2026 Waveframe Labs}