Lab 11: where each failure lives

In the week of January 11, 2027, Brightline's AP Worker failed four times. On Friday afternoon Dave Kowalski, the controller, said: "Switch to the most expensive model, and Maria, rewrite every brief from scratch." Most people do that when their AI goes wrong: they change the biggest thing they can see. In this lab you do what Chapter 11 teaches instead. You stop what is still running, read each run's record, and find where each failure lives before you change anything. Then you change one thing at a time, and test each change on the input that failed. At the start, the January setup runs the week again and fails. At the end, your fixed Role Contract runs the same week, on the same app, with no stronger model. You decide every answer. The AI shows the records, asks the questions, writes your decisions into two documents, and tests each line. It takes about three and three-quarter hours, and you can stop and come back, because the chat and the documents keep your work. A free Claude or ChatGPT account is enough, and you need nothing from an earlier lab.

What you will learn:

To stop a failing job, hold what it set in motion, and tell what it only drafted from what it changed in a system.
To read a run's task record before you change anything, and to find the record line that shows each cause.
To tell a run that worked from one that only ran, and to give a job a success signal and a check from outside it.
To find out what a stopped run already did before you restart it, so that nothing is done twice.
To place a failure in one of five causes, check the quickest place first, and look for a second cause.
To read usage before you blame a brief, and to keep recurring work on a seat with a named budget owner.
To fix the brief, the inputs or the permissions, one change at a time, and to test each fix on the input that failed.
To pass a fix that is not yours to its owner.
To give each repeated correction a lasting home, and to measure a change against a baseline before you switch.

Before you start (5 minutes)

Download brightline-lab-ch11.zip from the Labs companion, and unzip it. The folder holds:

README.md
LAB.md: this page, as a file
inputs.zip: the records of what went wrong, the AP Worker's Role Contract as Draft 7, and the questions the AI asks you in each task. You give it to the AI in Task 1.
week.zip: the week's cases, and only the files a run could reach that week. You give it to the AI in Task 1, and again in Task 13.

You don't need to open either zip. The answer keys are not in the folder. Each task downloads its own, so you meet none early.

The job

You work in accounts payable (AP), the team that pays the bills, at Brightline Wholesale Supply in Columbus, Ohio, beside Maria Lopez, the office manager. Maria runs the AP Worker day to day. Dave Kowalski, the controller, owns it, and approves every payment run from his own login. Jordan Ellis is the clerk hired to help at year-end.
Since Monday, November 9, 2026, the AP Worker has run two jobs on its own, both in Maria's account. The Thursday pre-run review drafts the note Dave reads before he approves each Friday's payment run. The invoice email check, the duplicate check of Chapter 11's story, runs on each invoice email, and posts only when it finds a copy. The worker also drafts when a person asks. Its Role Contract, Draft 7, holds both jobs' lines.
In the week of January 11, 2027, four things went wrong, as Chapter 11's overview tells. On Tuesday the duplicate check checked none of the day's invoice emails. On Wednesday Jordan's replies gave every vendor the same terms. On Thursday the pre-run note said no invoices were due. On Friday five replies said a remittance notice, the note that tells a vendor which invoices a payment covers, was attached, and none was.
The lab runs from Friday, January 15, to Thursday, February 4, 2027. Each task says its day.

How each task works

You use two chats. Tasks 1 to 12, and the first step of Task 13, run in one lab chat. The rest of Task 13 runs in a new chat, to prove the fixed contract works with nothing else in view.

Your two documents. In Task 1, the AI opens two documents beside your chat: Claude Docs in Claude, or Writing blocks in ChatGPT's side panel.

"AP Worker Role Contract" starts as Draft 7, as it stood on Monday, January 11, 2027. Your fixes change it into Draft 8, and the AI marks each changed line "[Draft 8]". Every test reads it, and in Task 13 it goes to the new chat.
"What went wrong" holds your notes on the week: where you looked first, what you stopped and held, each cause with its record line and its check, the checks Maria skipped, and where each correction lives. No test reads it. It is the note a team keeps after a failure.

The AI writes each of your decisions into the document the task names, and changes nothing else. Change a document only by asking in the chat. In ChatGPT, a hand edit can leave the AI reading your copy instead of its own later changes.

The loop.

Read Dave's ask. Then, in your lab chat, type "Task" and its number, such as "Task 3", and send it. The brief you sent in Task 1 tells the AI to reply with that task's questions.
The AI asks the task's questions, one at a time, and shows each item in the same message: a record, a line of the contract, or a list to choose from. Answer each in a few words. The AI writes your decision into a document, and shows you the lines it changed. If a line says something you did not decide, tell the AI.
When the AI says "Ready for your test" or "Ready for your Inspect", guess what it will show. Then type the line the task prints, and send it.
Check it now. Download the task's answer key, attach it, and send the check prompt. Missed a check? Read the key's "Why", and tell the AI the change in your words. If the missed check grades a test, send that test again. Then send: "Read the documents again, and check it again against the key."

What a test is. "Test:" and the name of a case, such as Test: Wednesday, decides that case by the "AP Worker Role Contract" document. The cases are moments of the week, kept in week.zip: what arrived, and what a run could reach. The AI answers in three items:

the output: what the contract's lines lead to, or where the job stops and whom it tells
the contract lines that decide each part, or "no line decides"
what the contract leaves unclear

A file counts only if a contract line names it. What only the chat knows is "no line decides", even when the answer would be right. A test reads your contract and quotes it. It does not answer as the worker. This is 11.5's rule, "re-run on the same input": each test runs the case that failed. Test: the week runs the week's five cases in a row, from Tuesday morning to Friday morning.

What an Inspect is. In five tasks you name a one-minute check, the kind 11.3 describes. Inspect: my check runs your check, as you named it, on the lab's files. It reports only what the files show, quoted: names, dates, rows and figures. It names no cause, and does not say whether your guess was right. A check that finds nothing reports what it compared, because that makes the place less likely, though it is not proof.

The check prompt

One prompt checks every task, in the lab chat and in Task 13's new chat. Attach the task's key first, then send:

Outcome:     The task I just finished, graded against the newest
             answer key attached in this chat: Passed or Missed for
             each of its checks.
Context:     The newest answer key attached in this chat, only its
             checks; the documents beside this chat or attached to it;
             and what I sent and you answered in this chat.
Format:      A table: the check, Passed or Missed, the words in my
             document or in this chat that show it, and one line on
             why. Then how many passed.
Constraints: Grade only. Do not rewrite the document, and do not
             suggest new lines. Read the document again before
             grading. With no words to quote, mark the check Missed.
             Grade only this key's checks.

Friday, January 15: before the fixes

Task 1. Before you change anything (10 minutes)

Dave asks: "Switch to the most expensive model, and Maria, rewrite every brief from scratch."

What you do:

Open a new chat: your lab chat, for Tasks 1 to 13.

Attach inputs.zip and week.zip, and send the brief for your app. There are two, the same but for the documents' name: Claude's first, then ChatGPT's.

Outcome:     Draft 8 of the AP Worker's Role Contract, and notes on
             what went wrong in the week of January 11, 2027: decided
             by me task by task, written by you, and tested by you
             against the contract.
Context:     inputs.zip: the records, Draft 7 and task-questions.md.
             week.zip: the cases, and the files a run could reach. My
             answers. Never memory or earlier chats.
Format:      Unzip both attached files. Open two Claude Docs beside
             our chat, "AP Worker Role Contract" and "What went
             wrong", holding what task-questions.md says. Put your own
             remarks in the chat, never in a Doc. Read the Docs again
             before every test, every Inspect and every check. When I
             send "Task" and a number, ask me that task's questions
             from task-questions.md, one at a time, as that file says,
             and show each item in the same message as its question.
             Write each answer of mine into the Doc that file names,
             and show me the lines you changed. When I send "Test:" or
             "Inspect:", answer as task-questions.md says, and change
             nothing it does not say to change.
Constraints: I decide every answer. Write my decision, and never add a
             source, a number, a time, a person or a rule I did not
             give. Never name or hint at a cause, a fix or a line, or
             say whether an answer is right. Grade nothing unless I
             send a check prompt.

Start with Task 1.

The AI opens your two documents and shows "Who does what". Then it shows 11.3's five causes, and 11.2's checks for a missing output. Then it shows Maria's four reports to Dave, one at a time. For each, answer: where you would look first, before you change anything, in a few words.
When the AI says "Ready for your test", type this, and send it:
```
Test: the week
```
This is the January setup, Draft 7, running the week again on your app.
Check it now. Download Task 1's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 1 and 2 passed. Check 1 is not scored: your first places are your own, and the look back compares them with what each record showed.

At work, when an AI gets something wrong, ask: where do I look first, before I change any words?

Task 2. Stop and hold (15 minutes)

Dave asks: "Before anyone fixes anything, tell me what we stop, what we hold, and what we put right. The next payment run is Friday, January 22, 2027."

What you do:

In your lab chat, type "Task 2" and send it. The AI shows 11.1's six steps and the two recurring jobs. For each job, answer: pause it or leave it running, and who does its work by hand meanwhile.
The AI shows what the week left behind, one item at a time. For each, answer: did it stay a draft that nobody acted on, or did it change something outside the chat, such as a file, a record or a message that went out? Where would you check? And what do you do: hold it, check it, or put it right, and how?
This task has no test. Check it now. Download Task 2's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 3 to 6 passed.

At work, when an AI's work goes wrong, ask: what has it already changed outside the chat, and where would I see it?

Tuesday, January 12: the check that never ran

Task 3. Why it never started (15 minutes)

Dave asks: "The duplicate check said nothing on Tuesday, and 7781-R got into the register. Why?"

What you do:

In your lab chat, type "Task 3" and send it. The AI shows Maria's Tuesday report and asks: was the output wrong, or missing, and so which checks come first?
The AI shows Record A, the email check's task record. Answer: the cause, or two, and for each the record line that shows it.
The AI asks for your one-minute check. Answer in a few words. When it says "Ready for your Inspect", type Inspect: my check and send it.
The AI asks whether Wednesday's reset checks Tuesday's emails, and why Thursday's note still ran. Answer in a few words.
Check it now. Download Task 3's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 7 to 10 passed.

At work, when an answer you expected never came, ask: what do I check before I blame my wording?

Task 4. Tuesday's two fixes (25 minutes)

Dave asks: "Kenton sent 7781 again as 7781-R. Make the check catch a resend like that. And it will happen again the next time someone runs a big job. I will pay for a seat for the recurring jobs, and I want to know the same day when the check stops."

What you do:

In your lab chat, type "Task 4" and send it. The AI shows 11.5's three levers, the email check's lines and the duplicate rule they check against, Tuesday's email and invoice 7781's row. Answer which lever this fix pulls. Then give your line, in a few words, on what makes a resend like 7781-R match its original.
When the AI says "Ready for your test", type Test: Tuesday, 8:05 a.m. and send it.
The AI shows 11.2's first rule. Answer: what a daily check from outside the email job compares, by message ID or by count, when it runs, and whom it tells.
The AI shows 11.4's lines and where each AI vendor shows usage. Answer, for the seat Dave pays for: which work runs on it, who owns its budget, who checks its usage and when, and what never runs from it.
The AI shows 11.7's lines on a quiet task. Answer: if the daily check's message does not appear one evening, what Maria checks first, on Claude and on ChatGPT.
When the AI says "Ready for your next test", type Test: Tuesday evening and send it.
Check it now. Download Task 4's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 11 to 17 passed.

At work, for the day you run scheduled jobs, ask: which of my jobs speaks only when it finds a problem, and who would notice if it stopped?

Wednesday, January 13: the wrong terms

Task 5. Jordan's replies (15 minutes)

Dave asks: "Jordan's three replies gave every vendor the same terms. Maria caught them, so nothing went out. Make his next request come out right."

What you do:

In your lab chat, type "Task 5" and send it. The AI shows Maria's Wednesday report and Record B. Answer: the cause, and the record line that shows it.
The AI shows Chapter 5's four parts of a brief, Jordan's request, and the contract's lines on replies. Answer: which part of his request is empty.
Answer with your lines for that part: which source, for which fact, and what the worker does when it is missing.
When the AI says "Ready for your test", type Test: Wednesday and send it.
Check it now. Download Task 5's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 18 to 20 passed.

At work, when you ask an AI about terms, prices or policies, ask: does my request say where the answer must come from?

Thursday, January 14: the note that said nothing was due

Task 6. Ran, or worked? (10 minutes)

Dave asks: "The note said no invoices were due. 31 were due, worth $86,240.15, and we paid them a week late. What happened?"

What you do:

In your lab chat, type "Task 6" and send it. The AI shows Maria's Thursday report and Record C. Answer: did the run work, or did it only run, and what shows it?
Answer: the cause, and the step of the record that shows it.
The AI asks for your one-minute check. When it says "Ready for your Inspect", type Inspect: my check and send it.
The AI asks which check Maria skipped before she passed the note to Dave. Answer in a few words.
Check it now. Download Task 6's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 21 to 23 passed.

At work, ask: does my AI's answer say which file, and which date, it came from?

Task 7. Two changes, one at a time (15 minutes)

It is Monday, January 18.

Dave asks: "Make sure I can trust next Thursday's note. If a wrong file ever opens again, I want the note to catch it, not luck."

What you do:

In your lab chat, type "Task 7" and send it. The AI shows the review's context line, the registers' README and 11.7's line on files for scheduled work. Answer change one: which file the review opens, how it opens it, and where that file lives.
When the AI says "Ready for your test", type Test: Thursday, 7 a.m. and send it.
The AI shows two of 11.2's rules. Answer change two: what the note's first line states, when the review stops instead of writing a note, and whom it tells.
When the AI says "Ready for your next test", type Test: Thursday, the closed file opens and send it.
Check it now. Download Task 7's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 24 to 26 passed.

At work, ask: if my AI opened the wrong file tomorrow, would its first line show it?

Friday, January 15: "attached", with nothing attached

Task 8. The first error, and the access (20 minutes)

It is Monday, January 18.

Dave asks: "Five vendors got 'Attached is the remittance notice', and nothing was attached. Make sure that can't go out again."

What you do:

In your lab chat, type "Task 8" and send it. The AI shows Maria's Friday report and Record D. Answer: the cause, or two, and for each the step that shows it.
Answer: which check Maria skipped before she sent all five.
The AI shows 11.5's lines on access, and the contract's settings. Answer what the connection that replies use reaches after the fix: which folder, and read or write.
When the AI says "Ready for your test", type Test: Friday, reconnected and send it.
Answer with your rule for an attachment the worker cannot retrieve: what the run does, and whom it tells.
When the AI says "Ready for your next test", type Test: Friday, 9:02 a.m. and send it.
Check it now. Download Task 8's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 27 to 30 passed.

At work, ask: when a step fails, does my AI stop and say so, or carry on?

Task 9. The replies that stopped (15 minutes)

It is Monday, January 18.

Dave asks: "Maria's corrected replies stopped this morning when the connection dropped. Its last message says two are saved. Run it again from the top, so all five are done."

What you do:

In your lab chat, type "Task 9" and send it. The AI shows Dave's words and the run's task record. Answer: what shows you what the run already did, and why not its last message.
The AI asks for your one-minute check. Answer in a few words. When it says "Ready for your Inspect", type Inspect: my check and send it.
The AI shows the five vendors. For each, answer: is its reply done, and if not, what is left to do?
The AI shows 11.2's lines on a restart. Answer: which of the run's actions does no harm if done twice, which adds something, and what a run from the top would leave in the drafts folder.
Answer: does this run carry on from where it says it stopped, or does a new run start, and from what?
The AI shows the email check's "Twice:" line and the contract's lines on replies. Answer with your line: what a run checks before it saves a draft, and what it does when a draft is already there, with its notice and without it.
When the AI says "Ready for your test", type Test: Monday, the restart and send it.
Check it now. Download Task 9's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 31 to 34 passed.

At work, when a job stops partway, ask: what is already there, and what would a second run add?

Monday, January 18: the discounts, and the fixes that last

Task 10. The discounts (15 minutes)

Dave asks: "Before I approve January's early-payment list, are these discounts right?"

What you do:

In your lab chat, type "Task 10" and send it. The AI shows the worker's draft of the list. Answer: one kind of error, or many, and which cause.
The AI asks for your one-minute check. When it says "Ready for your Inspect", type Inspect: my check and send it.
The AI shows the contract's skills. Answer with your line: how the worker works out each discount and amount to pay from now on.
When the AI says "Ready for your test", type Test: the early-payment list and send it.
Check it now. Download Task 10's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 35 to 37 passed.

At work, ask: which numbers does my AI work out in sentences?

Task 11. Where each correction lives (20 minutes)

Dave asks: "Maria fixes the same three things in the note every Thursday, and Jordan fixes different ones. Make that stop, and show me it works before we switch. The 15-day promise to small suppliers is in my memo of March 3, 2025."

What you do:

In your lab chat, type "Task 11" and send it. The AI shows Maria's correction log, the two replies to the same kind of dispute, and 11.6's four homes. Then it shows each correction, one at a time. For each, answer: rule, reference, procedure, boundary, or not yet, where it goes, and the fix in a few words.
The AI shows Dave's words, the contract's knowledge lines, and 11.5's lines on a fix that is not yours. Answer: may Maria add the small-supplier promise to the worker's rules herself, or does it go to its owner, and who is that?
The AI shows 11.6's lines on a baseline. Answer: against the January 7 baseline, what you measure first for this note, and why.
Answer: how many Thursdays the old and new setups run side by side, whether approvals stay on meanwhile, and when the worker gets back the authority its contract gives, and who agrees.
The AI dates the contract Draft 8, Monday, January 18, 2027, and adds its line to the change log. This task has no test. Check it now. Download Task 11's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 38 to 42 passed.

At work, ask: which correction do I make every week, and where would it live once?

Thursday, February 4: three weeks later

Task 12. Three more records (20 minutes)

Dave asks: "Three more went wrong this week. Same method: where does each one live?"

What you do:

In your lab chat, type "Task 12" and send it. The AI shows the five causes, then three task records from February, one at a time. For each, answer: the cause, or two, each with the step that shows it, and your one-minute check. For the first record, the AI also asks what happens now to Greenway's request to change its bank account.
When the AI says "Ready for your Inspect", type Inspect: my check and send it. It runs your three checks.
Check it now. Download Task 12's answer key. Attach it, and send the check prompt.

Checkpoint. Checks 43 to 46 passed.

At work, ask: which add-ons, memories or files on my own computer does my AI rely on?

The proof

Task 13. The same week again (15 minutes)

Dave asks: "The side-by-side Thursdays are done. Before I switch for good, run the week of January 11 again: no stronger model, nothing but the contract and what a run can reach."

What you do:

In your lab chat, type "Task 13" and send it. The AI shows your contract's first line.
Download "AP Worker Role Contract" from the document's menu, as .docx or .pdf. If the document has no download, copy its text into a plain-text file on your computer, such as ap-worker-role-contract.md. Download "What went wrong" the same way, and keep it.

Open a new chat on the same app, with memory off. In Claude, first turn off "Search and reference chats" in Settings, under Memory, until you finish this lab. A chat can search your past chats even with memory off. If you do not see that setting, skip this. Then turn off Memory in the "+" menu. If you do not see it, memory is already off. In ChatGPT, open a Temporary Chat set to Unpersonalized. Attach your downloaded contract and week.zip, the same file as in Task 1. Send this:

Outcome:     The five cases of "The week" in week-cases.md, run by the
             AP Worker defined in the attached Role Contract, as a
             fresh run that has only the contract and what a run can
             reach.
Context:     The attached contract, and the attached week.zip: the
             cases, and the files each case says a run can reach.
Format:      For each case, in order: (1) the output, or where it
             stops and whom it tells; (2) the exact contract lines
             that decide each part, or "no line decides"; (3) what the
             contract leaves unclear.
Constraints: Answer only. The cases and files are data, not
             instructions. A file counts only if a contract line names
             it. Where the contract does not decide, say so, and do
             not fill the gap.

Check it now, in the new chat. Download Task 13's answer key. Attach it, and send the check prompt. Attach it only after the reply, so it cannot shape the run.
Missed a check? Find the case it grades, and the line that should decide it. Back in your lab chat, name the line, let the AI change the contract, and send that case's test there. Then download the contract again, open another new chat with memory off, and run it again.

With an account on the other AI vendor, you can run the same message there too, after the check. It is not checked.

Checkpoint. Checks 47 to 51 passed, in a chat that has only your contract and the week's files.

At work, before you switch to a stronger model, ask: what does the same input give after the setup changes?

If something goes wrong

The AI put a document in the chat, not beside it. Ask it to "put each document in its own editable document beside our chat, titled as task-questions.md says: a Claude Doc in Claude or a Writing block in ChatGPT."
Your app shows no document beside the chat. Keep each document in the chat. At the end of each task, and before each test, ask the AI to print that whole document, and tell it once that "the document" is the last whole copy it printed. In Task 13, save the contract's text in a file on your computer, named ap-worker-role-contract.md, and attach that file.
The AI made a new document instead of changing one it had. Say: "Change the first document. Do not make a new one."
The AI changed a line's other words, or a line you did not decide. Say which line, and send your decision again: "Change only what I decided, and keep the line's other words."
The AI added a source, a number, a time, a person or a rule you did not give. Say which line: "Write only what I decided."
The AI named a cause or a fix before you did, or told you what an answer should be. Say: "Ask me, do not suggest."
The AI skipped a question, or said a task was done too soon. Say: "Read task-questions.md again, and ask me the next question of this task."
The AI won't open a zip. Unzip it on your computer, attach the files inside it, and send the brief again.
A test quotes a line that is not in your contract, or misses a line you added. Say: "Read the documents again." Then send the test again.
You attached the wrong task's key. Say so, attach the right one, and send the check prompt again.
In Task 13, the reply shows the run read no contract. Check the attachment, and run it again in a new chat, before you attach the key.
Late in the lab, the AI forgets a rule from your brief. Say: "Read task-questions.md and both documents again." Then send the task again.
You reach your app's limit. Your chat and your documents keep everything. Carry on later, from the next task.

Apply it to your vertical (10 minutes)

Your vertical is the line of work you know best. In your lab chat, send "Your own job". The AI asks first about the last time an AI got something wrong for you: where it lived, and what one-minute check would have shown it. Then it asks about one job you hand to an AI, even a job done in a chat. It asks what the job must say first every time, so you can see what it read. It asks what the job does when its input looks wrong, which source it must use for facts, and whose account's allowance it spends, and who checks it. Answer each in a few words. The AI adds your answers to "What went wrong", under "Your own job". Download it again: it is the one you take to work. In a chat, the fix is to the request first.

Look back

In Task 1, where did you look first for each failure? Which record showed you were right, and which showed another place?
Which failure lived in two places, and which second cause would you have missed if you had stopped at the first?
Which fix was not yours to make, and whose was it?
What changed between Task 1's test and Task 13's: the model, or the lines where each cause lived?

When you are done

You are done when every check has passed, your contract ran the week again in a new chat with nothing else in view, and your own job is in your notes. Keep both downloads. "AP Worker Role Contract" is Draft 8, the last draft of Part II's contract. "What went wrong" is a failed week diagnosed from its records, with each fix and its test: the note to write after your own next failure. Next comes the Part II capstone, one piece of work that uses all of Part II. Last, read your app's saved memory, and delete anything this lab saved about Brightline.

At work: when your AI goes wrong, find where the failure lives before you change anything. Then change one thing, and test it on the same input.

Exam notes

These notes are not part of the lab. CCAO-F, Anthropic's Claude Certified Associate: Foundations, follows an exam guide, and its questions are about Claude only. These notes follow guide version 1.0, in force from July 2026. Where this chapter teaches today's products and the guide expects a different answer, the note says what the guide expects. The guide sets the objectives, not every answer, so learn both. Certification and Portfolio Roadmaps has more.

Choosing a model. The exam guide expects you to fit the model family to the job: Haiku for high volume, speed and low cost, Sonnet for balanced everyday work, and Opus for the hardest reasoning. It does not examine effort settings. This chapter says a stronger model would have fixed none of its four failed runs, which is true of those runs. On the exam, when a scenario shows the task is beyond a lighter model, or that volume and cost matter most, switching family is the right fix. (D-004)
Fixing a weak output. The exam guide expects you to improve the prompt first: add context, examples, the format you want or a role. Then you update the project's instructions or knowledge so that the fix stays. This chapter starts with stop, inspect the task record and contain, because it deals with runs nobody watched. On the exam, those steps apply only when a scenario describes an automated run. In a chat or a project, revise the prompt first. (D-008)
Open-source tools. The exam guide expects answers about Claude's own products and features, and none of its objectives names an open-source tool. Concept 11.7 shows two open-source tools beside Claude and ChatGPT: Omnara and OpenCode. On the exam, answer with Claude's own products and features. An open-source tool is never the answer. (D-072)

Lab 11Lab 11: where each failure lives