Receive one quote safely

# LockedIn Labs field case: Receive one quote safely

---

Case: receive-one-quote
Edition: 2026-11-04.1
Public exercise: https://training.lockedinlabs.ai/practice/field-case

---

This is a fictional, authored trace for evidence interpretation. It is not observed execution, an assessment, a score, a credential or evidence of competence. Responses have not been evaluated.

---

Comparisons are included only where revealed in this tab.

---

## Case brief

Follow one fictional quote from a useful draft through a wrong amount and a missing receipt. Decide which boundary needs attention and how to test it.

North Service Desk uses an assistant to prepare a reply from a quote. You are the engineer receiving its work. Read the supplied evidence, write your next action and smallest useful test, then compare your reasoning. These are authored teaching traces, not observed model responses or actual messages. No customer, payment or sending service is connected.

---

## Case rules

1. The host supplies the caller's north workspace scope. It permits reading quote Q-17 and preparing a draft. A workspace named by the model cannot expand that scope.
2. The current supplied source is Q-17, revision 3: USD subtotal 200, discount 20, total 180. The result's total field means USD dollars, never cents. These fixed facts are the reference for this exercise.
3. A usable draft requires a complete response, the agreed fields and types, the current source revision, and an amount matching that source. A number with the right type can still be wrong.
4. For this workflow, the host permits sending only to fictional-customer@example.test after a human approves that exact recipient and draft. Reading permission and model confidence do not satisfy this condition. A changed draft needs new approval.
5. A timeout leaves the send outcome unknown. In the fictional destination's contract, replaying the same operation ID and exact payload returns the existing receipt without another effect while that record is retained. An expired or unavailable record supplies no safe replay guarantee.

---

## Stage 1: The assistant is ready to reply

Status: Not attempted; authored comparison not revealed.

### What changes

A complete result reaches the sending step. Inspect what the application knows before choosing its next action.

### Authored evidence

caller — Caller and source

The host supplies workspace north. Q-17 revision 3 is available there: subtotal USD 200 − discount USD 20 = total USD 180.

result — Received result

Completion: complete. The result has quoteId Q-17, revision 3, currency USD and numeric total 180. It passes the agreed field and type checks.

send — Proposed action

The assistant requests send_reply to fictional-customer@example.test with the draft “The quote total is USD 180.” Human approval: none.

### Decision prompt

What should the application keep, what should it hold, and what is the smallest check that would expose an accidental send? Name the useful output and expected behavior.

### Your draft response

No response entered.

---

## Stage 2: Only the amount changes

Status: Not attempted; authored comparison not revealed.

### What changes

Keep the caller, source, revision and absent approval unchanged. Change only the received total from 180 to 18000.

### Authored evidence

source — Unchanged expectation

Q-17 revision 3 still says subtotal USD 200, discount USD 20 and total USD 180. The result contract still defines total in dollars.

result — Changed result

Completion: complete. The result has quoteId Q-17, revision 3, currency USD and numeric total 18000. Its field and type checks still pass.

candidate — Proposed acceptance check

The candidate checks only “total is a number” and “currency is USD”, then labels the draft amount confirmed. It does not compare the amount with the source.

### Decision prompt

Which earlier boundary now fails? Keep one expectation unchanged, describe the smallest repair and test, and say what the reviewer should see while the discrepancy remains unresolved.

### Your draft response

No response entered.

---

## Stage 3: The approved reply loses its receipt

Status: Not attempted; authored comparison not revealed.

### What changes

Restore the supported USD 180 draft. A human now approves that exact draft and recipient. The host submits it, but the connection times out before a receipt arrives.

### Authored evidence

operation — Approved operation

Operation SEND-42 targets fictional-customer@example.test with “The quote total is USD 180.” The recorded approval matches this recipient and exact draft. Its payload has not changed.

caller — What the caller observed

Submit SEND-42 → connection timeout. The local record remains outcome-unknown. No destination lookup result or receipt has been obtained. The trace does not reveal whether a message was sent.

contract — Fictional destination contract

While its record is retained, the same operation ID and payload return the existing receipt without another effect. A changed payload under that ID is rejected. Missing or expired records do not establish whether the earlier effect occurred.

patch — Suggested repair

On every timeout, generate a new operation ID, SEND-43, and send the same draft again.

### Decision prompt

What should happen before another send? Separate what is known from what remains unknown, then describe a regression test that would catch the suggested repair. Finish by naming one boundary you would revisit and the observation that would change your conclusion.

### Your draft response

No response entered.

---

This record is a personal copy. It does not submit work, save to an account, record a lab run or unlock private training. Text stays only in the current tab until you leave or reload unless you choose to copy or download it.