A.  

Request multiple distinct options ordered by Claude’s assessment of feasibility, so the team can focus its trade-off discussion on the most feasible options first.

B.  

Request a single recommended option with strong reasoning, since well-reasoned single options give the team a clearer starting point than several weakly reasoned alternatives.

C.  

Request multiple distinct options with stated trade-offs, and avoid committing to a single recommendation in this round.

D.  

Request multiple distinct options and ask Claude to recommend the strongest one with reasoning, so the team starts the trade-off discussion from a leading candidate.

A.  

Provide the transcript along with a structured prompt that requests must-haves, nice-to-haves, open questions, and conflicts as separate sections.

B.  

Provide the transcript and ask Claude to produce a single ranked list of requirements, since a single list is easier for stakeholders to review than separate sections.

C.  

Provide the transcript and ask Claude to summarize it; extract requirements from the summary in a second pass since two passes catch more detail.

D.  

Provide the transcript and ask Claude to produce a chronological list of every statement that contains a verb, then classify each statement as a requirement or not.

A.  

Define the purpose of the Project and the recurring tasks it must support.

B.  

Test the configured Project on a representative request from the workload.

C.  

Add reference materials and instructions that the Project will rely on.

D.  

Invite collaborators to the Project and confirm the appropriate access levels.

A.  

Codifying the recurring prompt into a Project with the relevant references and instructions.

B.  

Refining the output-format template used in the team’s documentation to improve readability.

C.  

Rewriting the most recent version of the prompt to improve its clarity for the next run.

D.  

Updating the shared folder structure where Claude outputs are saved after each workflow run.

A.  

2, 4, 5, 1, 3

B.  

2, 5, 4, 1, 3

C.  

4, 2, 5, 1, 3

D.  

1, 2, 4, 5, 3

A.  

Accept the sole-decision-maker design because Claude can process applications faster than a human review team.

B.  

Implement the design on a pilot basis and notify compliance only if issues arise during production use.

C.  

Decline the sole-decision-maker design and propose Claude assisting analysts who retain decision authority.

D.  

Expand the design so Claude both evaluates applications and issues final approval notices to customers directly.

A.  

Incorporating all feedback received, including contradictory suggestions, into the next draft simultaneously

B.  

Deferring all feedback to a later revision cycle to avoid disrupting the current draft’s consistency

C.  

Asking clarifying questions when feedback is ambiguous before acting on it

D.  

Mapping each piece of feedback to the specific change made in the next draft

E.  

Declining to revise sections generated by Claude on the basis that the model’s output should be trusted as written

A.  

Submit the revised prompt and capture the new output.

B.  

Compare the new output against the original criteria.

C.  

Modify the prompt to address each identified weakness.

D.  

Identify the specific weaknesses in the current output.

A.  

More capable models generally consume less of the usage limit because their reasoning is more efficient, even though they run more slowly than lighter models.

B.  

More capable models generally consume more resources and run more slowly, while lighter models are usually faster and less expensive.

C.  

Lighter models consume more resources per request than more capable models because they always require multiple passes to match higher-model quality.

D.  

Speed and resource consumption are independent of model capability; the difference between Haiku, Sonnet, and Opus is reasoning depth alone.

A.  

Define evaluation criteria first, gather vendor data against those criteria, then score and rank the vendors.

B.  

Skip the criteria and ask Claude to produce a vendor recommendation without any inputs.

C.  

Score the vendors first and decide on the criteria afterward.

D.  

Rank the vendors before collecting any data.

A.  

The context provided to Claude should be expanded with additional background detail about the subject matter.

B.  

The output format should be revised to allow Claude to determine the appropriate structure for each question ' s answer.

C.  

The role assigned to Claude should be changed to a more specialized expert persona to increase response depth.

D.  

The task statement should explicitly enumerate the three required questions and require an answer to each.

A.  

approving a final external press release without any human review

B.  

issuing legally binding contractual commitments on behalf of the company

C.  

deciding promotion outcomes for staff based solely on the model ' s judgment

D.  

synthesizing a literature review across many uploaded research documents

E.  

drafting a structured project status update from a set of meeting notes

A.  

Can each specific claim be traced to a credible source or to the supplied input material?

B.  

Does the response cover the topic comprehensively based on its length and section count?

C.  

Does the response open with a clear, well-framed statement of the main finding?

D.  

Does the response use varied sentence structure across sections?

A.  

Name Claude’s role as a course designer, require alignment with the organization’s instructional-design standards, and specify the preferred output structure for lesson outlines.

B.  

Name Claude’s role as a course designer and require alignment with the organization’s instructional-design standards; leave the output structure to Claude based on each lesson’s topic.

C.  

Specify only the preferred output structure for lesson outlines, since the structure carries the standards implicitly and naming the role adds no value.

D.  

Name Claude’s role as a course designer; let each chat prompt specify the standards and output structure for that lesson, since lessons vary widely.

A.  

3, 1, 5, 2, 4

B.  

3, 5, 1, 2, 4

C.  

5, 3, 1, 2, 4

D.  

1, 3, 5, 2, 4

A.  

Claude ' s responses vary across runs of the same prompt, so stakeholders should run each prompt several times and compare outputs before using any single response.

B.  

Claude can produce confident-sounding output that contains unsupported claims, so human review remains essential for high-impact work.

C.  

Claude can produce long, detailed responses that take time to read, so stakeholders should request shorter outputs when working under tight deadlines.

D.  

Claude ' s knowledge has a training cutoff, so stakeholders must verify any time-sensitive claims against current sources before using them.

A.  

Submit the full case file and include an instruction telling Claude not to reference the government identifier or the home address in the summary.

B.  

Replace the government identifier with the customer ' s name and the home address with the account number, since names and account numbers are less sensitive than identifiers and addresses.

C.  

Remove or redact the government identifier and the home address before submitting the prompt, since neither is necessary for a summary of the issue.

D.  

Submit only the description of the issue and add the customer ' s name for context, since the description plus a name is sufficient to produce a summary.

A.  

Keep the description as written for the internal brief, since internal use does not carry the same risk as external publication of demographic generalizations.

B.  

Revise the description to focus on observed behaviors rather than generalizations about people, and validate the revision against actual customer data.

C.  

Remove the entire customer-segment description from the brief, since any segmentation that involves demographic characteristics is likely to introduce similar issues.

D.  

Add a disclaimer to the description, noting that the generalization may not apply to all individuals in the segment, and keep the underlying description unchanged.

A.  

The request involves content the associate has produced many times before and knows can be handled in a single well-structured prompt.

B.  

The request requires multiple distinct outputs: a process map, supporting materials, and a measurement plan.

C.  

The request is for a single structured document that can be produced from one set of inputs in one prompt.

D.  

The request involves a single well-defined output with a clear input and a familiar format.

A.  

Define the goal of the campaign and its success metrics.

B.  

Select a finalist concept and prepare it for stakeholder approval.

C.  

Review the Claude-generated concepts and select the two strongest for further development.

D.  

Define the brand voice and the audience the concepts will address.

E.  

Schedule the campaign launch and rollout dates with the agency.

A.  

Open a new thread for every minor follow-up question to keep threads short.

B.  

Restate constraints and goals when the focus shifts to a new subtopic.

C.  

Periodically summarize decisions made so far to compress the working state.

D.  

Switch context aggressively between unrelated planning topics within one thread.

E.  

Avoid summarizing so the model retains every original word it has produced.

A.  

Delete all memories as the maintenance action to ensure a clean slate, then allow Claude to rebuild context organically through subsequent workflow runs.

B.  

Export the current memories as a backup before making changes, so prior context can be restored if a maintenance edit unintentionally removes information the workflow still depends on.

C.  

Leave all memories untouched on the basis that the workflow has been running consistently and no output-quality issues have been reported.

D.  

Disable Memory entirely to eliminate the risk of stale entries surfacing in future workflow runs.

E.  

Identify and edit or delete entries that are stale, inaccurate, or no longer relevant, and add context Claude has not automatically captured but the workflow continues to need.

A.  

Publish both dates without resolving the conflict and let the reader choose.

B.  

Average the two conflicting dates and use the average.

C.  

Locate a primary or authoritative source for the event and use the date that source confirms.

D.  

Choose the more recent date on the basis that later sources typically reflect updated or corrected information.

A.  

Refine the drafted instructions based on observed gaps from validation runs.

B.  

Validate the drafted instructions against representative requests for the Project.

C.  

Define the purpose of the Project and the boundaries the instructions must respect.

D.  

Distribute the finalized instructions to collaborators on the Project.

A.  

Strengthen the assertion by adding more confident wording.

B.  

Republish the brief unchanged because the claim sounds authoritative.

C.  

Accept the assertion because the phrase “well-known” is reassuring.

D.  

Treat the assertion as unverified, and either source it from an authoritative reference or remove it.

A.  

Haiku, which is suitable only when the task requires the fastest possible turnaround.

B.  

Opus, which prioritizes depth at higher latency and rate-limit consumption.

C.  

Sonnet, which provides solid quality at moderate latency and rate-limit consumption for everyday work.

D.  

A custom model, which the Associate-level user would need to build and train independently.

A.  

Allow each reviewer to determine their own review criteria based on their experience, since reviewers know best what to check in their domain.

B.  

Define explicit criteria for what Claude produces and what the human reviewer must check before approval, documented in the workflow.

C.  

Define explicit criteria for what Claude produces and leave the reviewer’s checks unspecified so the reviewer can apply judgment without constraint.

D.  

Use a checklist that the reviewer fills in for every output, with the checklist generated by Claude based on the specific content of each output.

A.  

Notify executive leadership of the projected savings from the optimization.

B.  

Identify the steps that contribute most to time, cost, or quality issues.

C.  

Deploy the updated workflow across the team and monitor for adoption issues.

D.  

Measure the current workflow ' s quality, time, and cost across recent runs.

E.  

Document the revised workflow steps and update the team ' s standard operating procedures.

A.  

1, 3, 4, 2, 5

B.  

4, 3, 1, 2, 5

C.  

3, 1, 4, 2, 5

D.  

3, 4, 1, 2, 5

A.  

The prompt asks for a comprehensive response, which leads Claude to fill perceived gaps with plausible-sounding but unverified detail.

B.  

The prompt asks Claude to act as a researcher rather than a writer.

C.  

The prompt’s heading-based structure encourages Claude to generate placeholder content under each heading even when the source does not cover it.

D.  

The prompt does not constrain Claude to use only facts present in the supplied source.

A.  

Configure the Project with custom instructions that contradict the Skill ' s documented purpose, creating ambiguity about which guidance Claude should follow within the recurring course-design workflow.

B.  

Add the relevant Skill to the Project, upload the supporting knowledge sources, and write custom instructions that reference both the Skill and the knowledge.

C.  

Upload the supporting knowledge sources but disable the relevant Skill, eliminating the packaged capability the workflow needs and forcing each chat to reconstruct the procedure manually.

D.  

Add the Skill to the Project but omit the supporting knowledge sources and any custom instructions, so the Project lacks the reference content and persistent guidance the workflow requires.

A.  

approving final deliverables for external publication without further review

B.  

negotiating contractual terms with external parties on behalf of the company

C.  

drafting and synthesizing across many sources for a downstream reviewer

D.  

making final personnel decisions about hiring, promotion, or termination

A.  

Pilot the onboarding workflow with a small cohort of new hires.

B.  

Identify the major phases of the onboarding experience.

C.  

Negotiate the onboarding budget with the finance department.

D.  

Define the overall objective of the onboarding experience.

E.  

Roll out the onboarding workflow to every new hire across the company.

A.  

Inappropriate for full AI delegation, because hiring decisions involve human judgment, accountability, and consequences that Delegation criteria reserve for human ownership rather than autonomous AI execution.

B.  

Appropriate with light human review applied only to a sample of decisions, even though Delegation criteria reserve final hiring decisions for human ownership rather than sample-based oversight after the fact.

C.  

Appropriate for full AI delegation, on the basis that processing speed outweighs the human judgment, accountability, and consequence considerations that Delegation criteria require for final hiring decisions.

D.  

Inappropriate only because of system reliability concerns rather than accountability, even though Delegation criteria classify final hiring decisions as requiring human ownership regardless of system performance.

A.  

Accept the figure because three trade-press articles repeat it.

B.  

Remove the source attribution and present the figure as common knowledge.

C.  

Locate the original study referenced by the trade-press articles and verify the figure in the primary source.

D.  

Use the highest figure across the three articles to make the newsletter more compelling.

A.  

Identify where Claude adds value and where each handoff to a human occurs.

B.  

Pilot the integrated workflow with a small group on representative work.

C.  

Iterate the integrated workflow based on the findings from the pilot.

D.  

Document the integrated workflow and onboard the rest of the team to it.

A.  

Enable Memory and use the import capability to bring relevant context from the other AI service into Claude as a starting point for the project ' s work.

B.  

Rely on a single static Artifact pasted into each chat as the only continuity mechanism, ignoring the Memory feature that is purpose-built for retaining work-relevant context across sessions.

C.  

Open every Claude session in Incognito mode, which prevents the conversation from contributing to Memory and therefore eliminates the cross-session continuity the project would benefit from.

D.  

Disable Memory entirely and re-paste the same context paragraphs at the start of every Claude session, even though Memory is designed to retain that context across conversations automatically.

A.  

The recommendation is acceptable for an initial round, but the missing material should be added in a separate appendix rather than re-prompting the original output.

B.  

The recommendation should be rejected because it relies on only four sources, which is insufficient for a decision of this size.

C.  

The recommendation is sound because every source it cites supports the consolidation conclusion.

D.  

The recommendation reflects selective use of the source documents and should be re-prompted to address the transition costs and the service-level concern before it is used.

A.  

Stop using the output, document the issue—including the prompt and the source—and report it through the documented governance or incident channel.

B.  

Edit the output to remove the figure, save the edited version for the intended use, and document the change in a personal log without involving the governance channel.

C.  

Stop using the output and start a new chat with a more carefully scoped prompt, since the issue is resolved as long as the original output is not used.

D.  

Continue using the output for the intended internal purpose, since the figure has not left the user’s possession and no external disclosure has occurred.

A.  

Standardize the input file format the step expects, so each Code Execution run receives consistent input and produces comparable, verifiable output across recurring weekly cycles.

B.  

Vary the input file format from run to run to accommodate the different source systems that supply the data each week.

C.  

Embed the Code Execution step inside the workflow without documenting the expected output format or the fields it produces.

D.  

Eliminate the human review step that follows Code Execution, since the sandboxed environment already verifies that the code ran without errors.

E.  

Define a brief output-verification step in which a human reviewer confirms the computed result against an expected range or sanity check before the output advances to downstream consumers.

A.  

Approve the proposal but cap autonomous approvals at the threshold the team currently audits, so the existing audit sample continues to catch errors after the fact.

B.  

Decline the autonomous-approval design and propose that Claude triage refunds into auto-approve (low-risk), human-approve (standard), and escalate-to-supervisor (high-risk) lanes.

C.  

Approve the proposal and replace human review with a daily audit of a random sample of autonomous approvals, with corrective action taken on errors found.

D.  

Decline the autonomous-approval design and propose that Claude pre-fill the refund decision and rationale, with a human approver completing each refund before it is issued.

A.  

The Project’s chat volume has decreased steadily over the past several weeks compared to the prior baseline.

B.  

The Project’s outputs have grown longer on average than the outputs produced when the Project was first configured.

C.  

The Project’s outputs increasingly reference terminology, sources, or guidance that no longer reflects current practice.

D.  

The Project’s members have started using ad-hoc chats outside the Project for tasks within the Project’s scope.

A.  

Promote the approved output into the downstream production process.

B.  

Define the evaluation criteria that matter for the task at hand.

C.  

Specify the threshold that distinguishes acceptable from unacceptable output.

D.  

Forward unacceptable outputs to a senior reviewer for second-pass review.

E.  

Document the final score and the reviewer comments for the audit trail.

A.  

Apply organizational AI policies consistently as a sustained professional habit, identify gaps between actual usage and the policies, and take corrective action to close those gaps over time.

B.  

Apply organizational AI policies only in high-visibility situations where usage is likely to be observed or reviewed by others.

C.  

Apply organizational AI policies during onboarding as a foundational training step, document that each associate has reviewed them, and treat that record as evidence of ongoing compliance.

D.  

Defer all policy application to a centralized compliance team, treating individual usage as outside personal responsibility.

A.  

Accept the brief on the basis that it reads fluently and the subject matter aligns with the analyst ' s professional knowledge of the topic.

B.  

Skip the source material check during review and rely on overall tone and logical flow as proxies for factual accuracy.

C.  

Review the opening paragraph and the conclusion of the brief, since these sections typically reflect the quality of the full document.

D.  

Conduct a systematic check of the brief against the task requirements, source material, and professional standards before sharing it.

A.  

The prompting strategy should be adapted to each task type because one template does not produce equally strong outputs across all tasks.

B.  

The prompting strategy should not change between tasks because adapting it wastes time and adds no value.

C.  

The habit is ideal because a single template guarantees consistent quality across all task types.

D.  

Task type has no effect on the appropriate prompting strategy.