Domain 7 – Troubleshooting and Optimisation (10%)

• Diagnose underperforming prompts or poor outputs
• Adjust approach based on feedback and results
• Optimise personal and team workflows for efficiency

CCAO-F Domain 7 · Task 7.1

What You Need to Know

Diagnosing means naming why a reasonable-looking request produced a poor output, before changing anything.

The Routing Question

Did that happen once, or does it happen every time?

• Once → prompt problem (better brief)
• Every time from the same Project → configuration problem (fix the instructions or knowledge once)

Six Causes (Cover Nearly Everything)

1. Context starvation (by far the most common)
2. Instruction overload
3. A configuration defect
4. A degraded conversation
5. Stale or superseded material
6. A feature or setting mismatch

Starvation: supply what was missing. Asking more forcefully supplies nothing. Overload: break into substeps, group and prioritise.

When the Conversation Itself Is the Problem

A long chat accumulates superseded drafts and unrelated topics. Fix: start a fresh conversation carrying a verified summary of what still matters. Projects handle large recurring material through retrieval.

Failure Modes Anthropic Names

• Fabricated quotes that look right → scrutinise high-stakes claims; review cited sources
• Hallucinated capability (working-looking links that fail, claim to have sent an email) → Claude has no access to tools not integrated
• Too brief → retool the prompt; give Claude the previous prompt and response
• Error messages (capacity, length-limit) → retry, or chunk/summarise/start fresh — not reword the prompt

The Escalation Reflex

Reaching for the most capable model because the output disappointed is the domain’s favourite trap. A vague brief fails at every tier. Reach for feature/setting changes only after the brief is sound.

Key Concept: Classify the symptom before reading the options: once or every time, and which named failure mode. The keyed answer fixes the cause.

Exam Traps

• Rewriting today’s prompt when every output has the same defect
• Escalating to a more capable model or higher effort as the first response to a disappointing output
• Treating an error message as a quality defect and rewording the prompt
• Chasing the delivery of an email Claude claims to have sent in a chat with no email integration

Practice Scenario

Art gallery administrator drafts loan-request letters inside a correspondence Project. For two months, every letter has come out in American spelling; she has corrected each one by hand. A more capable model gave the same spelling. She wants to stop the hand corrections and keep drafting inside the Project. Most appropriate next step?

Build Exercise: Diagnose Your Last Disappointment (25 min)

CCAO-F Domain 7 · Task 7.2

What You Need to Know

An adjustment round means changing a request, running it again and judging the output. Three parts make it systematic:

1. Change One Thing at a Time

One change at a time shows what fixed the output. A bundled change looks efficient until you need a second good result — and nobody knows which of the three worked.

2. Hold the Test Input Fixed

A new prompt tried on a new document tells you nothing, because two things moved. Holding the test input fixed makes round-to-round comparison meaningful. Editing an earlier message creates a different version of the chat with its own artifacts.

3. State the Criterion Before You Look

“Better” is not a criterion. “A customer can find the total price in five seconds” is. Stating it in advance stops your judgement drifting towards whatever the new draft happens to do well.

Feedback Arrives from Readers, Not Just the Text

A colleague says the tone is off. A manager returns a draft. A reader asks the same clarifying question three times. These usually locate the gap more precisely than re-reading the text yourself.

Know When to Stop Adjusting

When several well-aimed rounds have not moved the output, the assumption that the approach is right has failed. The third failed adjustment is a signal to go back to 7.1’s cause list and re-diagnose. A fourth round of slightly different words tests the same failed assumption.

Exam Traps

• The shotgun fix dressed as decisiveness
• A fourth round of rewording after three aimed rounds moved nothing
• Testing the adjusted prompt on a different document and crediting the prompt

Practice Scenario

Ceramics studio owner drafts the joining email she sends attendees before each weekend workshop; attendees keep saying it is too long and too formal. She has time for one adjustment round this week. Which round is set up to teach her something?

Build Exercise: Run One Honest Adjustment Round (25 min)

CCAO-F Domain 7 · Task 7.3

What You Need to Know

Overhead is anything you type, explain or attach at the start of every similar conversation. Optimising means making your work with Claude cost less time or usage. Measure first.

The Repeated-Overhead Test

Recurring overhead belongs somewhere it only has to exist once:

• Standing context and rules → Project’s instructions
• Recurring reference material → Project knowledge
• Near-identical weekly request → saved form you adapt per use
• Preferences that apply to everything → account-wide personalisation settings

Documented Levers

• Batching — group multiple related tasks in a single message rather than fragmenting them
• File names — well-named files help Claude retrieve the right information (“Worktop price list” beats “pl_v3_final”)
• Match settings to the work — routine reformatting at maximum effort pays for depth it never uses; deep analysis at a low setting underserves the task

The Proportion Rule

Some overhead is not worth removing. A twenty-minute task done once a year does not earn a Project, a saved form or an afternoon of setup. Recognising the optimisation that costs more than it saves is as professional as spotting the one that pays.

Key Concept: Look for the measurement. The keyed answer moves measured, recurring overhead to where it exists once, or declines a setup that costs more than it saves.

Exam Traps

• Sweeping workflow changes proposed off a feeling
• Building a Project or saved form for a task done twice a year
• Raising effort or switching model to speed up a workflow whose overhead is a repeated preamble

Practice Scenario

City farm development officer starts every funder conversation the same way: explaining what the farm is and which programmes it runs, then attaching the same three documents. Timed: about ten minutes, three or four times a week. Each January she also spends twenty minutes preparing the annual return for the charity regulator. What is the most appropriate optimisation?

Build Exercise: Time the Overhead, Then Place It or Leave It (30 min, spread over a week)