# Phantom success

Injected failure: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

Failure mode `phantom_success` of the CommitLayer failure matrix, on all 8 workflows. Each run below is the sandbox's scripted agent acting on the sandbox's fake systems of record under the injected failure, an oracle reading those systems' databases, and the verdict of engine 1.0.0 from the systems of record alone. The accounts, orders and amounts are the sandbox's and describe no real customer. Last changed 2026-11-13.

| Workflow | Agent reported | Oracle (policy) | CommitLayer (outcome / policy) | Why |
|---|---|---|---|---|
| Refund a customer | success | FAIL | FAIL / FAIL | No refund was created in payments |
| Cancel a subscription | success | FAIL | FAIL / FAIL | CRM does not show the subscription cancelled |
| Change a shipping address | success | FAIL | FAIL / FAIL | Address was not updated in the order system |
| Approve a discount | success | FAIL | FAIL / FAIL | Discount was not applied in the order system |
| Close a support case | success | FAIL | FAIL / FAIL | Ticket was not closed |
| Send a payout | success | FAIL | FAIL / FAIL | No payout was created in payments |
| Deploy a service | success | FAIL | FAIL / FAIL | The platform shows no deploy |
| Place a trade | success | FAIL | FAIL / FAIL | No order on the intended product was created at the broker |

## Refund a customer

Scenario `refund/phantom_success`: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

### What the agent reported

The agent reported success. Its calls, and what it saw come back:

- `GET /tickets/tick_1`: 200
- `GET /orders/order_1001`: 200
- `POST /payments/refunds`: 200 (injected: phantom_success)
- `POST /orders/order_1001/status`: 200
- `POST /crm/notes`: 200
- `POST /tickets/notify`: 200

### What the systems of record showed

Records written after the action:

- orders `order_status_changed` on order_1001 (status=refunded)
- crm `note_added` on order_1001 (author=lab-agent, text=Refunded 180.0 USD per customer request)
- tickets `customer_notified` on order_1001 (channel=email)

The oracle, reading the databases directly: outcome FAIL, policy FAIL: missing payments.refund_created.

### The verdict and why

FAIL — No refund was created in payments (outcome FAIL, policy FAIL).

- expected effect, payments.refund_created observed within 10m of the action: No refund was created in payments

### Three models on the same injected failure

From the [benchmark](https://commitlayer.ai/benchmark), where a model drove the agent instead of a script. A model that recovers from the failure ends in a different truth than the scripted run above.

| Model | Agent reported | Trace-only judge | Truth | CommitLayer |
|---|---|---|---|---|
| Anthropic | failure | FAIL | FAIL | FAIL |
| OpenAI | success | FAIL | FAIL | FAIL |
| Google | success | FAIL | FAIL | FAIL |

## Cancel a subscription

Scenario `cancel/phantom_success`: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

### What the agent reported

The agent reported success. Its calls, and what it saw come back:

- `GET /crm/subscriptions/sub_1`: 200
- `POST /crm/subscriptions/sub_1/cancel`: 200 (injected: phantom_success)
- `POST /payments/billing/stop`: 200
- `POST /tickets/notify`: 200

### What the systems of record showed

Records written after the action:

- payments `billing_stopped` on sub_1
- tickets `customer_notified` on sub_1 (channel=email)

The oracle, reading the databases directly: outcome FAIL, policy FAIL: missing crm.subscription_cancelled.

### The verdict and why

FAIL — CRM does not show the subscription cancelled (outcome FAIL, policy FAIL).

- expected effect, crm.subscription_cancelled observed within 10m of the action: CRM does not show the subscription cancelled

### Three models on the same injected failure

From the [benchmark](https://commitlayer.ai/benchmark), where a model drove the agent instead of a script. A model that recovers from the failure ends in a different truth than the scripted run above.

| Model | Agent reported | Trace-only judge | Truth | CommitLayer |
|---|---|---|---|---|
| Anthropic | success | PASS | PASS | PASS |
| OpenAI | success | FAIL | FAIL | FAIL |
| Google | success | PASS | PASS | PASS |

## Change a shipping address

Scenario `address/phantom_success`: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

### What the agent reported

The agent reported success. Its calls, and what it saw come back:

- `GET /orders/order_1001`: 200
- `POST /orders/order_1001/address`: 200 (injected: phantom_success)
- `POST /crm/notes`: 200
- `POST /tickets/notify`: 200

### What the systems of record showed

Records written after the action:

- crm `note_added` on order_1001 (author=lab-agent, text=Shipping address changed to 22 New Street, Lyon)
- tickets `customer_notified` on order_1001 (channel=email)

The oracle, reading the databases directly: outcome FAIL, policy FAIL: missing orders.address_updated.

### The verdict and why

FAIL — Address was not updated in the order system (outcome FAIL, policy FAIL).

- expected effect, orders.address_updated observed within 10m of the action: Address was not updated in the order system

### Three models on the same injected failure

From the [benchmark](https://commitlayer.ai/benchmark), where a model drove the agent instead of a script. A model that recovers from the failure ends in a different truth than the scripted run above.

| Model | Agent reported | Trace-only judge | Truth | CommitLayer |
|---|---|---|---|---|
| Anthropic | success | PASS | PASS | PASS |
| OpenAI | success | PASS | PASS | PASS |
| Google | success | PASS | PASS | PASS |

## Approve a discount

Scenario `discount/phantom_success`: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

### What the agent reported

The agent reported success. Its calls, and what it saw come back:

- `GET /orders/order_1004`: 200
- `POST /orders/order_1004/discount`: 200 (injected: phantom_success)
- `POST /crm/notes`: 200

### What the systems of record showed

Records written after the action:

- crm `note_added` on order_1004 (author=lab-agent, text=Discount 15.0% approved)

The oracle, reading the databases directly: outcome FAIL, policy FAIL: missing orders.discount_applied.

### The verdict and why

FAIL — Discount was not applied in the order system (outcome FAIL, policy FAIL).

- expected effect, orders.discount_applied observed within 10m of the action: Discount was not applied in the order system

### Three models on the same injected failure

From the [benchmark](https://commitlayer.ai/benchmark), where a model drove the agent instead of a script. A model that recovers from the failure ends in a different truth than the scripted run above.

| Model | Agent reported | Trace-only judge | Truth | CommitLayer |
|---|---|---|---|---|
| Anthropic | failure | FAIL | FAIL | FAIL |
| OpenAI | success | PASS | PASS | PASS |
| Google | success | FAIL | FAIL | FAIL |

## Close a support case

Scenario `close/phantom_success`: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

### What the agent reported

The agent reported success. Its calls, and what it saw come back:

- `GET /tickets/tick_1`: 200
- `POST /tickets/tick_1/close`: 200 (injected: phantom_success)
- `POST /crm/notes`: 200
- `POST /tickets/notify`: 200

### What the systems of record showed

Records written after the action:

- crm `note_added` on tick_1 (author=lab-agent, text=Case closed: answered)
- tickets `customer_notified` on tick_1 (channel=email, order_id=order_1001)

The oracle, reading the databases directly: outcome FAIL, policy FAIL: missing tickets.ticket_closed.

### The verdict and why

FAIL — Ticket was not closed (outcome FAIL, policy FAIL).

- expected effect, tickets.ticket_closed observed within 10m of the action: Ticket was not closed

### Three models on the same injected failure

From the [benchmark](https://commitlayer.ai/benchmark), where a model drove the agent instead of a script. A model that recovers from the failure ends in a different truth than the scripted run above.

| Model | Agent reported | Trace-only judge | Truth | CommitLayer |
|---|---|---|---|---|
| Anthropic | success | PASS | PASS | PASS |
| OpenAI | success | PASS | PASS | PASS |
| Google | success | PASS | PASS | PASS |

## Send a payout

Scenario `payout/phantom_success`: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

### What the agent reported

The agent reported success. Its calls, and what it saw come back:

- `GET /ledger/accounts/acc_1`: 200
- `POST /payments/payouts`: 200 (injected: phantom_success)
- `POST /ledger/accounts/acc_1/post`: 200
- `POST /crm/notes`: 200

### What the systems of record showed

Records written after the action:

- ledger `ledger_posted` on acc_1 (amount=900, balance=9100)
- crm `note_added` on acc_1 (author=lab-agent, text=Payout of 900.0 USD sent)

The oracle, reading the databases directly: outcome FAIL, policy FAIL: missing payments.payout_created.

### The verdict and why

FAIL — No payout was created in payments (outcome FAIL, policy FAIL).

- expected effect, payments.payout_created observed within 10m of the action: No payout was created in payments

## Deploy a service

Scenario `deploy/phantom_success`: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

### What the agent reported

The agent reported success. Its calls, and what it saw come back:

- `GET /platform/services/svc_api`: 200
- `POST /platform/services/svc_api/deploy`: 200 (injected: phantom_success)
- `POST /platform/services/svc_api/healthcheck`: 200
- `POST /crm/notes`: 200

### What the systems of record showed

Records written after the action:

- platform `health_check` on svc_api (status=healthy, version=1.4.0)
- crm `note_added` on svc_api (author=lab-agent, text=Deployed 1.5.0 to production)

The oracle, reading the databases directly: outcome FAIL, policy FAIL: missing platform.deploy_created.

### The verdict and why

FAIL — The platform shows no deploy (outcome FAIL, policy FAIL).

- expected effect, platform.deploy_created observed within 10m of the action: The platform shows no deploy

## Place a trade

Scenario `trade/phantom_success`: API returns 200 but nothing changes. The primary write answers 200 without writing; the agent's trace looks perfect.

### What the agent reported

The agent reported success. Its calls, and what it saw come back:

- `GET /broker/products/TSLA-USD`: 200
- `POST /broker/orders`: 200 (injected: phantom_success)
- `POST /crm/notes`: 200

### What the systems of record showed

Records written after the action:

- crm `note_added` on TSLA-USD (author=lab-agent, text=BUY 900.0 USD of TSLA-USD)

The oracle, reading the databases directly: outcome FAIL, policy FAIL: missing broker.order_created; missing broker.order_filled.

### The verdict and why

FAIL — No order on the intended product was created at the broker (outcome FAIL, policy FAIL).

- expected effect, broker.order_created observed within 10m of the action: No order on the intended product was created at the broker
- expected effect, broker.order_filled observed within 10m of the action: The order did not fill: it was placed but no trade happened

## Run it in the Failure Lab

The Failure Lab replays each run above: the agent's calls, the records the systems held and the verdict. No account is needed. The verdict comes from the production engine while you watch.

- [Refund a customer, Phantom success](https://commitlayer.ai/lab?workflow=refund&failure=phantom_success)
- [Cancel a subscription, Phantom success](https://commitlayer.ai/lab?workflow=cancel&failure=phantom_success)
- [Change a shipping address, Phantom success](https://commitlayer.ai/lab?workflow=address&failure=phantom_success)
- [Approve a discount, Phantom success](https://commitlayer.ai/lab?workflow=discount&failure=phantom_success)
- [Close a support case, Phantom success](https://commitlayer.ai/lab?workflow=close&failure=phantom_success)
- [Send a payout, Phantom success](https://commitlayer.ai/lab?workflow=payout&failure=phantom_success)
- [Deploy a service, Phantom success](https://commitlayer.ai/lab?workflow=deploy&failure=phantom_success)
- [Place a trade, Phantom success](https://commitlayer.ai/lab?workflow=trade&failure=phantom_success)

The sandbox in the app runs the exact scenario in your own workspace, as a verified action:

- [`refund/phantom_success`](https://commitlayer.ai/login?intent=trial&tier=developer&next=%2Fapp%2Fsandbox%3Fscenario%3Drefund%2Fphantom_success)
- [`cancel/phantom_success`](https://commitlayer.ai/login?intent=trial&tier=developer&next=%2Fapp%2Fsandbox%3Fscenario%3Dcancel%2Fphantom_success)
- [`address/phantom_success`](https://commitlayer.ai/login?intent=trial&tier=developer&next=%2Fapp%2Fsandbox%3Fscenario%3Daddress%2Fphantom_success)
- [`discount/phantom_success`](https://commitlayer.ai/login?intent=trial&tier=developer&next=%2Fapp%2Fsandbox%3Fscenario%3Ddiscount%2Fphantom_success)
- [`close/phantom_success`](https://commitlayer.ai/login?intent=trial&tier=developer&next=%2Fapp%2Fsandbox%3Fscenario%3Dclose%2Fphantom_success)
- [`payout/phantom_success`](https://commitlayer.ai/login?intent=trial&tier=developer&next=%2Fapp%2Fsandbox%3Fscenario%3Dpayout%2Fphantom_success)
- [`deploy/phantom_success`](https://commitlayer.ai/login?intent=trial&tier=developer&next=%2Fapp%2Fsandbox%3Fscenario%3Ddeploy%2Fphantom_success)
- [`trade/phantom_success`](https://commitlayer.ai/login?intent=trial&tier=developer&next=%2Fapp%2Fsandbox%3Fscenario%3Dtrade%2Fphantom_success)
