CCAO-F practice questions, answered and explained

An employee supplies a policy document and asks a question about it. The answer is fluent and cites a section number. What is the most reliable way to check it?

• A Open the cited section in the supplied document and read it

Correct. The document is in hand, so the citation can simply be checked — direct, conclusive, and quick.

• B Ask the same question in a new conversation and compare the two answers

Agreement between two runs is not evidence: both draw on the same tendencies and can be consistently wrong.

• C Ask the model how confident it is in the answer

A confidence statement is produced the same way the answer was, so it inherits the same uncertainty rather than measuring it.

• D Judge by the fluency and internal consistency of the answer

Fluency and internal consistency are properties of the writing and hold just as well for an incorrect answer.

The source is available, so verification is a matter of reading it — the cheapest and most conclusive check there is. Fluency is unrelated to accuracy. Two conversations can agree and both be wrong, since they share the same tendencies. A stated confidence level is generated text, not a measurement of correctness.

Which feedback is most likely to produce a useful revision?

• A "Try again, and this time produce something rather different from that one entirely please"

This produces a different attempt rather than a corrected one.

• B "Make it rather better than this, to a standard the reader will genuinely appreciate reading"

"Better" supplies no direction to move in.

• C "This isn't quite right, so please have another go at it and see whether you can do any better"

It reports dissatisfaction without saying what is wrong or where.

• D "The second section assumes the reader knows the process; explain it in one sentence first"

Correct. It locates the problem, names the cause and states the remedy.

Useful feedback names what is wrong, where, and what would fix it. The others describe dissatisfaction without locating it, so the revision is a guess. Specificity is what makes the next attempt targeted rather than random.

Which step in a process is the best candidate to keep with a person?

• A The most repetitive step, since repetition is tiring and demands attention

Repetitive steps are the natural candidates to hand over.

• B The longest step, since it consumes the largest share of the time

Length is a reason to want help rather than to withhold it.

• C The first step, since everything after it depends on getting it right

Position in the sequence says nothing about consequence.

• D The step where a judgement call commits the organization to something

Correct. A commitment needs someone accountable for it, whatever answer could be generated.

Commitment is where accountability matters and where an error is expensive to undo — so that step stays with someone who can be answerable for it. Repetition and duration are reasons to hand a step over, and position in the sequence is not a criterion.

What belongs in a Project's knowledge rather than in an individual question?

• A The name of whoever requested the piece of work

The requester changes per task and rarely affects the content.

• B Reference material that most questions in that Project will need

Correct. Material most questions need is what makes shared knowledge worth having.

• C The deadline by which the whole of the work has to be finished in the end

Deadlines vary per task and do not shape the answer.

• D Today's specific question, so that it is available to everyone else later

Today's question is the definition of task-specific.

Project knowledge holds what recurs across the work. Material most questions draw on belongs there so it does not have to be supplied each time. Questions, deadlines and requesters vary per task and belong with the task.

What is the practical value of an organization having a stated policy on assistant use?

• A It is legally required everywhere, so an organization without one is straightforwardly in breach of the law

Requirements vary; a policy is good practice rather than a universal legal mandate.

• B It replaces individual guesswork with a common standard, so decisions are consistent and defensible

Correct. It replaces individual judgement with a common, defensible standard.

• C It reduces the cost of the tools, since having a stated policy supports better licensing terms overall

Policy does not affect what the tools cost.

• D It improves output quality, since a stated policy sets a standard the output must always meet

Quality comes from how tools are used rather than from the existence of a policy.

Without a policy every employee decides for themselves what is acceptable, producing inconsistent practice and no defensible position when something goes wrong. A stated standard makes decisions consistent and reviewable. It does not change cost or quality, and it is not universally mandated.

An employee asks for "talking points for the client call." The result reads as a script rather than points. What was underspecified?

• A How the material will be used — read aloud, glanced at, or handed over

Correct. How the material is used is what determines its form.

• B The call's duration, which determines how many points there is time for

Duration bounds how much, not what shape it takes.

• C The employee's role on the call, which shapes what they should be saying

The employee's role shapes perspective rather than format.

• D The client's name, without which the points cannot be made specific

The client's name affects content, not whether points or prose are wanted.

"Talking points" describes a form loosely; how they will be used determines it precisely. Points to glance at while speaking look nothing like a script to read or a leave-behind. Naming the use settles the form. Client name, duration and role affect content rather than shape.

An output covers everything asked but stops short of a recommendation. What determines whether this is a shortfall?

• A Whether a recommendation was part of the ask, or whether the decision is the requester's to make

Correct. It depends on the ask and on whose decision it is.

• B The length of the output, since a fuller answer would surely have reached a recommendation of its own

Length says nothing about whether a recommendation was owed.

• C Whether the topic is contentious, since a contentious one plainly warrants a good deal more caution

Contentiousness may be a reason to withhold rather than to supply.

• D Whether the analysis is strong enough to be able to carry a recommendation on its own at all

Strong analysis can still correctly stop short of deciding.

Sometimes withholding a recommendation is correct — the decision belongs to someone with context or authority the analysis does not have. Whether it is a gap depends on what was asked and whose call it is. Length, analysis quality and contentiousness do not settle it.

Why can an assistant give a good answer about a well-known topic and a poor one about a niche internal process?

• A The question about the internal process was probably worse, and a better one would have worked perfectly fine here

The question quality is not what differs between the two cases.

• B Internal processes are usually badly designed, so a clear answer about one is genuinely difficult to produce here at all

Process quality is unrelated to whether it can be described.

• C Well-known topics are well represented in training data; an internal process is not represented at all unless supplied

Correct. Public topics are represented in training data; internal processes are not.

• D Niche topics are inherently harder to reason about, whatever material happens to be available about them anywhere at all

The reasoning is the same; the material available differs.

The difference is availability. A public topic is extensively represented; a company's internal process appears nowhere unless it is supplied. The reasoning capability is the same; the material is not.

What does supplying a file's context — what it is and why it matters — most improve?

• A Extraction accuracy, since context makes the individual figures easier to locate

Extraction is a mechanical step that framing does not change.

• B The file's formatting, since context establishes how the material should be laid out

Formatting is a property of the file.

• C The relevance of the answer, since the same document supports very different questions

Correct. The same document supports different answers depending on why it was supplied.

• D Processing speed, since supplying context tells the assistant where to look and what it can skip

Speed is unaffected by contextual framing.

A document can be read for many purposes, and the same text supports different answers depending on why it was supplied. Saying what it is and what matters about it directs the reading. Speed, extraction and formatting are unaffected.

An employee wants a colleague to be able to reproduce their result later. What matters most?

• A Using the same model, since a different one will not reproduce the same result reliably

Model consistency helps but the inputs matter more.

• B Saving the output itself, so the colleague can see exactly what the result was supposed to look like

Saving the output preserves a result rather than the ability to reproduce it.

• C Recording the time at which the work was done, so the result can be placed against whatever was current then

Timing is irrelevant.

• D That the material and instructions live somewhere shared rather than in one person's conversation

Correct. Reproduction requires the inputs, which must be shared.

Reproduction requires the inputs, and inputs held in a private conversation are unavailable to anyone else. Shared material and instructions make it possible. Saving the output preserves a result rather than the ability to reproduce it.

An employee reviews output by reading it once for anything obviously wrong. What class of error does this miss?

• A Spelling errors, which a single reading passes over because the eye supplies the intended word

Spelling errors are visible on a read.

• B Plausible-sounding errors, which are precisely the ones that survive a read for obvious problems

Correct. A read for obvious problems misses exactly the errors that look right.

• C Missing sections, which are invisible on a single pass because there is nothing on the page to notice

Missing sections are noticeable against the ask.

• D Formatting errors, which a reader following the argument stops noticing after the first page

Formatting problems are visible.

A scan for obvious problems catches what looks wrong, and the dangerous errors are the ones that look right. Catching those requires checking against a source or a requirement rather than reading for impressions. Spelling, formatting and missing sections are all visible on a read.

An employee wants every output to include a specific disclaimer. Where does this belong?

• A In each request, so that the disclaimer is stated freshly every single time it is needed

Per-request statements depend on remembering every time.

• B In the document template only, since that is where the finished text is assembled

A template covers documents but not other outputs.

• C Added manually afterwards, so a person confirms it on every single output

Manual addition is the rework the instruction removes.

• D In standing instructions, since it applies to every output regardless of the task

Correct. A universal requirement belongs stated once.

A requirement that applies universally belongs stated once. Per-request statements and manual addition both depend on remembering, which is what fails. A document template covers documents but not other outputs.

An employee is asked to redact a document before supplying it. What is the most common mistake?

• A Redacting the wrong file and supplying the unredacted original by mistake

File confusion is a process error rather than a redaction one.

• B Redacting far too much, so the document loses the detail the task actually needed

Over-redaction reduces usefulness but is not the common failure.

• C Removing names while leaving detail that identifies the person by combination

Correct. Combinations of remaining detail frequently identify the person.

• D Using the wrong color of marking, so the redaction is not obvious to a reader

Presentation is not the substantive issue.

Identification does not require a name. A role, a date, a location and a specific history together frequently single someone out, particularly in a small organization. Effective redaction considers the combination rather than a single field.

When is pasting a long source document into the prompt the wrong move?

• A Whenever the document is longer than a few thousand words in total

Length alone is not disqualifying when the whole document bears on the question.

• B When only a defined part of it bears on the question being asked

Correct. Supplying the irrelevant remainder dilutes attention across it.

• C When the document is already stored somewhere the team can reach

Where a document is stored says nothing about whether it belongs in context.

• D When the document contains formatting that will not survive pasting

Lost formatting is a nuisance and rarely changes the substance of an answer.

Relevance decides this, not length. A long document that is wholly relevant belongs in the prompt; a long document of which one section matters dilutes attention across the rest. Storage location, formatting loss and raw length are practical considerations rather than reasons not to supply what the question needs.

Which is the strongest evidence that an output's tone fits its audience?

• A The model confirms the tone matches the audience that was specified

Self-assessment repeats the instruction rather than testing it.

• B It avoids jargon, contractions and any informal constructions

That describes one register, which may or may not be the fitting one.

• C It resembles documents that audience has responded well to before

Correct. Demonstrated response from that audience is the evidence that counts.

• D The employee who requested it finds the tone appropriate on reading

The requester is a proxy for the audience and frequently a poor one.

Fit is a claim about a particular audience, so the best available evidence is that audience's demonstrated response to comparable material. The requester is a proxy and often not a member of the audience, self-assessment restates the instruction, and avoiding jargon and contractions is one style rather than evidence of fit.

An employee wants to know whether a cheaper model is good enough for a drafting task. What is the sound test?

• A Ask the more capable model to assess the cheaper model's output

A model's assessment of another's output is not independent evidence.

• B Run both on the same real inputs and compare against the criteria

Correct. Both models, real inputs, stated criteria: that is the measurement.

• C Try the cheaper model once and see whether the output looks acceptable

One sample tells you about one sample.

• D Read the published capability descriptions for the two models

Published descriptions are general and cannot settle a particular task.

Good enough is a claim about this task, so it is settled by running both on representative inputs and judging against the criteria that define a good result. Published descriptions generalise across tasks, a single sample is not a measurement, and one model's opinion of another is not evidence.

What should a reusable template make explicit that a one-off prompt can leave implicit?

• A The name of the person who wrote the template

Authorship supports questions later but does not enable use now.

• B The reasoning behind the way the prompt is constructed

Rationale helps maintainers and does not tell a user what to supply.

• C The inputs the user is expected to supply and where

Correct. A stranger to the prompt needs the fill-in points marked.

• D The model that the template was originally written against

Model provenance is metadata and rarely affects use.

A template is used by people who did not write it, so the parts they must fill in have to be marked. Rationale is useful documentation, and authorship and model provenance are metadata; none of them is what makes the template usable by someone else.

An employee asks for a summary of candidate feedback and the output emphasises different qualities for different candidates. What should they do?

• A Ask the model to apply the same criteria to every candidate

Re-prompting acts before knowing whether anything is actually wrong.

• B Accept it, since the feedback about each candidate differed

That may be true, and it has not been established.

• C Check the emphasis against the source before relying on it

Correct. Only the source shows whether the emphasis was faithful or introduced.

• D Rewrite the summary to give each candidate equal length

Equal length is cosmetic symmetry rather than accuracy.

Differing emphasis may faithfully reflect differing feedback or may be introduced by the summarization, and only the source distinguishes them. Accepting it assumes the first, re-prompting acts before the cause is known, and equal length imposes symmetry without establishing accuracy.

Removing half the supplied context makes the output better. What is the most likely explanation?

• A The removed material was formatted in a way that confused the model

Formatting problems rarely resolve this cleanly on removal.

• B The remaining material was more recent than what was removed

Recency carries no weight of its own within supplied context.

• C The removed material was contradicting the rest of the context

Correct. Improvement on removal points at active interference, usually contradiction.

• D The shorter context is processed more carefully throughout

Less material competes less, but that is dilution rather than care.

Improvement on removal points at the removed material actively interfering, most often by contradicting what remains. Recency is not weighted, shorter context is not read more carefully as such, and formatting rarely produces a clean quality improvement when removed.

An employee spends twenty minutes crafting a prompt for a task that would take fifteen minutes by hand. When is this justified?

• A When the prompt will be reused on similar tasks afterwards

Correct. The cost amortises across every subsequent use of the prompt.

• B It is not, since the prompt cost more than the task

That compares a reusable asset against a single use of it.

• C When the employee learns something about prompting from it

Learning is a genuine return, though an incidental one.

• D When the resulting output is of higher quality than manual work

A quality gain is a real benefit and a smaller one than repeated reuse.

A prompt is an asset, so its cost is amortised across every future use and a one-off comparison misprices it. Quality gains and learning are real benefits but they are secondary to reuse, and judging it purely on the first run ignores that the prompt persists.