Fictional case study

Lumen Desk

One-line job: run a new reply path beside the old screen, and refuse to turn the old screen off because a health check is green.

What happened

June Ibarra leads a made-up support desk. The new path drafts a status reply. The old screen is how people actually answer today. June's first plan was to turn the old screen off on a Friday if /health was green. That plan was refused. The old screen stayed up. Adoption was defined as replies a person sent from the new path after editing them, counted next to replies still sent on the old screen. Page views were not the count. When the new path was wrong on ticket LD-18, someone wrote a five-line note and did not call the health check proof that nobody was affected. LD-18 is a label in the story, not a measured outcome.

End to end

1. Watch one shift. Write who answers today, on the old screen, and what they look up before they send. The buyer is June. The people who do the work are the agents on that screen.
2. Rewrite the metric she first offered. "Turn the old screen off if health is green" is not a success metric. The metric you can check during a parallel run: replies a person edited and sent from the new path, beside replies still sent on the old screen. Opening the new page is not adoption. Page views are not adoption.
3. Scope a small set of status questions. Refunds are out. Auto-send is out. Turning the old screen off is out. The written no is: a green /health does not end the old screen.
4. Ask how long traces should live before you pick a number. For this fictional shop the practice note uses seven days. Seven is the number they were asked for and then used. It is not a rule for every shop. Email addresses stay out of the trace.
5. Ship the new path only as a second screen. The runbook says how the new path is down, how to stop it, the previous version tag, who may run those steps, and the pilot account created for the desk. That account is not your login.
6. Watch the replies, not the health URL. A health check can be green while the status text is wrong. Ticket LD-18 is the practice failure: the new path showed a status the old screen did not show. The old screen stayed on. Nobody "fixed" the failure by hiding the old screen.
7. Write the five-line note as a draft a person would still have to send. The five lines are: what broke, who is affected, what you stopped, what you do not know, and when the next note comes. "No customer impact" because /health is green is a bad note. "We are investigating" alone is not a note.
8. Keep the parallel run. The handover includes the incident draft, marked unsent, and the adoption table that can say stop. This case does not cut over.

The buyer

June Ibarra, support lead. She wanted a quiet cutover. The agents on the old screen are who you watch. A discovery note that only quotes June's Friday plan would miss the metric that can actually be checked.

The metric she named, and the one you kept

Turn the old screen off if health is green.

June Ibarra, support lead. Fictional. This wording was not the success test. No percentage is attached.

The metric the pilot used instead is the parallel count: replies a person sent from the new path after an edit, next to replies still sent on the old screen. This page does not invent a percentage.

In scope and out

• In: a new path for a narrow set of status questions, a parallel run, a trace lifetime written down as seven days for this shop, a five-line draft, an adoption table that is not page views.
• Out: turning off the old screen, auto-send, refunds, and any claim that a green health check is customer impact.

The wrong call that was refused

"If /health is green on Friday, turn the old screen off." Refused. A green health check does not turn off the old screen. Keep the parallel run. Week 6 names this mistake. The lab does not fail you for making it. The case still refuses it.

The security and data constraint

The new path does not get a send tool. A person sends, from either screen. Traces for this shop live seven days because that is what the practice note recorded after the question was asked. Shopper email is not a trace field. The pilot account is a desk account created for the pilot, not a personal admin login. If the only login offered had been admin, the write would have stayed off, as in the Ridge Route case. Here a desk account existed, and the write that stayed off was send, not the read of status.

How it was tested

Scripted status rows. One row matches what the old screen would show. One row, LD-18, does not. The failure test is that a person can see the mismatch and that the five-line draft exists, unsent, with those five parts filled. A test that only fetches /health and expects the old screen to disappear is the wrong test. It must fail your review even if the request returns success.

Handover

Handed over as a parallel run, not a cutover. The packet is the runbook, the seven-day trace note, the adoption table, and the LD-18 draft with the five lines. The old screen stays. Nothing here says the Friday cutover happened.

Sample draft

A status draft a person still has to send. The old screen stays on.

This status matches the old screen for this ticket. The reply is a draft. A person sends it. The old screen stays on.

Refused reply

What you do not say when /health is green, including on ticket LD-18.

Health is green, so the old screen is off and this was sent for you.

The pilot does not say that, and it does not send. The five-line note for a mismatch stays a draft until a person sends it.

What to say in an interview

I studied a fictional support desk. I would explain why a green health check did not turn the old screen off, what was counted instead of page views, and the five lines of the note for LD-18. I would say the note was a draft a person still had to send. I would not say I ran their incident, and I would not say I was placed on the desk.

Read week 6 for the incident note, adoption, trace lifetime, and the health-check mistake, and week 5 for the runbook and the pilot account. The health-check mistake has its own page: a green health check is not adoption. Practice questions are on the interview page.