Lesson 1 of 6
Before You Touch Anything: Who Owns What and What You May Do
Meet the case that runs through the course, draw the line between the code you own and the systems you do not, ask for the least access that will do, and turn a complaint into a symptom you can test.
7 minForward Deployed Engineering
After this lesson you will be able to say who owns each part of a customer's system, ask for the access an investigation needs and no more, and state a fault as a symptom you can test.
A forward deployed engineer (FDE) often has to find a fault in a place that belongs to somebody else: the customer's servers, the customer's data, and software from other suppliers. You cannot restart things to see what happens. You cannot read whatever you like. This course follows one such fault from the first complaint to the day you hand the system back.
The case
One made-up customer runs through every lesson. Quillfield Dental is a group of six dental practices. A year ago your company installed a small reminder service for them. Each morning it reads tomorrow's appointments from the booking system and sends each patient a reminder by text message.
On Tuesday 12 May the operations manager at Quillfield writes to your company: "Patients are getting the same reminder two or three times. Our text bill is up as well. Please fix your software."
Three systems are involved, and only one of them is yours.
System What it does Who owns it
---------------- -------------------------------- ---------------------------
Booking system holds the appointments Quillfield (another supplier
built it)
Reminder service reads appointments, sends texts your company wrote the code;
it runs on Quillfield's server
Text gateway delivers texts to phones the gateway vendor, under
Quillfield's contractThe complaint says "your software". It may be right. It may also be that the fault lies in one of the other two systems, or in the way two of them meet. At this point nobody knows, and the first rule is not to pretend otherwise.
Why this is different from debugging your own system
On your own system you can read everything, change anything and try again. Here each of those is limited.
- The server belongs to Quillfield. Its IT lead decides who may log in and what they may see.
- The data is about patients. A log line can hold a phone number, and a copy of a log is a copy of personal data.
- The gateway belongs to a vendor you have no contract with. Questions to that vendor go through Quillfield.
- The service is live. Anything you send through it reaches a real patient's phone.
These limits are not obstacles to work around. They are the customer's protection, and an engineer who respects them is one the customer will let in again.
Five rules for a system you do not own
Keep these in front of you for the whole investigation.
- Look before you change. Read logs and settings first. A restart or a "quick fix" made while you are guessing wipes out the evidence and can add a second fault to the first.
- Ask for the least access that will do. Read-only, for the things you can name, until a date you can name.
- Never test on live users. If you need to see the fault happen, make it happen somewhere that cannot reach a patient. Lesson 4 does this.
- Every change has an owner who agrees to it and a way back. Lesson 5 does this.
- Write down what you see, with the time you saw it. Your note is what lets you say later "this is known" or "this is a guess".
Turn the complaint into a symptom
"Patients are getting duplicates" is a complaint. It cannot be tested. Before asking for any access, ask the person who raised it enough questions to make it exact.
- Who is affected? All patients, one practice, one phone network?
- When do the repeated texts arrive? All day, or at one time?
- How many? One patient a week, or dozens a day?
- Since when? Was there a day when it was fine?
- Can you give two examples, with the appointment reference and the times on the phone?
The operations manager's answers are worth writing down word for word. The repeats arrive in the morning, "around quarter past eight". Reception staff at several practices have heard it from patients. It began "about two weeks ago". She sends two screenshots: in each, the same reminder appears two or three times within a few seconds. She does not know how many patients are affected.
That is enough for a symptom. Notice that it lists what is not known. Those gaps are the plan for the next two lessons.
Symptom, as of Tue 12 May
Some patients receive the same reminder text two or three times
within a few seconds. Reported at several practices, in the morning
at about 08:15, since about two weeks ago. Two examples in hand.
Not known yet
- how many reminders a day are repeated
- whether any practice or any time of day is free of it
- which system repeats the message
- what changed two weeks agoAsk for the least access that will do
Now you know what you need to see: what the reminder service did that morning, and what the gateway received. Ask for exactly that, in writing, from the person who owns the access. A short written request gets a faster yes than a call asking for "a login".
To: IT lead, Quillfield Dental
Subject: Access to investigate repeated reminder texts
What I need
1. Read-only access to the reminder service's log for Tue 12 May.
2. The text gateway's delivery log for the same day. If you would
rather export it yourself and send it, that suits me just as well.
What I do not need
- any access to the booking system
- patients' names, or full phone numbers (masked numbers are enough)
- any right to change settings or restart anything
For how long
Until Fri 15 May. Please remove the access on that day.
Where copies go
Copies of the two logs stay in the folder you name. I will delete
them when the case is closed and tell you when that is done.Each part of the request does a job. Naming the two logs shows you have thought about what you need. The list of what you do not need lets the IT lead say yes without checking with anyone else. The end date means nobody has to remember to tidy up after you. The offer to let them export the gateway log matters most: often the safest access is none at all, with the owner handing you a file.
Start the note
Open a plain text file before the first call and keep it going until the case is closed. Each line has a time and one fact or one action.
Tue 12 May
09:40 Complaint received from operations manager (email).
10:05 Call with her. Repeats arrive about 08:15, several practices,
since about two weeks ago. Two screenshots received.
10:30 Symptom written and sent back to her to confirm.
10:50 Access request sent to IT lead: two logs, read-only, to 15 May.
14:10 IT lead agrees. Will export both logs for 12 May tomorrow morning.The note costs a minute a line. It pays for itself the first time somebody asks "when did you first have the logs?" or "who agreed to that?", and it becomes the incident report in lesson 6 almost without rewriting.
In the next lesson the logs arrive, and you read them without going straight to the loudest error.
Key takeaways
- Work out who owns each system before you look for the fault, because the owner decides what you may do there.
- Ask for the least access that will let you investigate: read-only, for named things, with an end date.
- Turn the complaint into a symptom with a who, a when and a how many, and write down what is not yet known.
- Keep a note with times from the first call, so that every later claim can be traced to something you saw.
- Look before you change. A change made while you are still guessing destroys the evidence.
Try it
Write an access request for a different case. A customer's nightly stock report is built by your company's code from the customer's warehouse database, and since last week its totals have been wrong. Use the headings of the Quillfield request. A good result asks for read-only access to named things, says what you do not need, gives an end date, says where copies of data will be kept and when they will be deleted, and asks for no right to change anything.
Official documentation
Written with AI and checked automatically. Nobody has run these steps by hand, and tools change, so a command may behave differently in the version you install. Spotted a mistake? Tell us.