CCAO-F exam prep Reading the questions, and doing the work
Decision tools that work across domains
Four questions that settle items in several domains at once, each drawn from the guide rather than invented as an exam trick
· Part 3 · Reading the questions, and doing the work · 3 min read
The seven domains are not really separate. One item can touch two or three. This lesson gathers the criteria the guide scatters through its sections into questions you can ask yourself.
Every criterion here comes from the guide, not from exam technique.
One — has this gone past what an Associate does
Section 1 makes this part of the role.
"recognizing limitations and escalating more complex or technical work to Claude Architects and Developers"
and Section 3 draws the line.
the Associate's productivity work, process design, use cases
not the Associate's coding against the API, agentic system design,
enterprise-scale AI architectureWhen an item reaches the right-hand column, the right answer is usually to escalate, not to attempt it.
Two — does a human have to check this
Domain 2 states it outright.
"Determine when human review or additional verification is required"
The first answer key supplies two usable triggers.
a specific factual claim a subsection, a figure, a date, a name
a recipient who cannot compliance, legal, finance, a customer
absorb an errorThe guide stresses "especially citations bound for a compliance audience".
And it rules out the model's confidence as a trigger. Options that have the model assess itself are wrong in this shape.
Three — can this data go in
Domain 6 gives a clear order in the third answer key.
1 policy stated? if so, that policy is the rule
2 identifiers? if so, redact or anonymise first
3 still works? if so, do it, do not abandonStep three is the one that matters most, because the guide marks abandoning a task that anonymisation would have enabled as "unnecessary".
Four — what decides the model
Domain 3 gives three axes, and the second answer key gives a usable rule.
high volume, straightforward, speed matters faster, cheaper model
complex reasoning the most capable modelThe guide's phrase is "reserving the most capable model for complex reasoning".
"Reserving" means holding it back for when it is needed, not making it the default.
How much the four questions cover
question 1 escalate? domains 4 and 6
question 2 human review? domains 2 and 7
question 3 data allowed? domains 5 and 6
question 4 which model? domains 3 and 7Together they touch every domain except 1, which measures the writing of prompts more than the making of decisions.
What this does not prove
The four questions are not in the guide in this form. Each criterion is, but grouping them this way is my arrangement. The guide proposes no such decision order.
And they do not cover everything. An item testing feature knowledge directly — the difference between an artifact and an inline answer, say — is not helped by any of the four.
References
Official documents
- CCAO-F Exam Guide Section 1 on escalation, Section 3 on scope, Section 6 for the objectives, Section 8 for the answer keys
Computed here
- Gathering the criteria into four questions, and the table of which question touches which domain, is this lesson's arrangement rather than the guide's