Pass Anthropic CCA-F Exam in First Attempt Easily

Prompt instructions cannot enforce authorization of irreversible side effects. This alternative suggests “Rely on a reminder to check identity before using the refund tool”, whereas the supported approach is “Gate the refund tool in application code on a verified identity state”. A prompt reminder cannot guarantee that an external money-moving action observes required order. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Let the refund tool run and verify identity afterward” is the wrong action for this case. The prerequisite must be satisfied before money moves. The supported choice is “Gate the refund tool in application code on a verified identity state”. A prompt reminder cannot guarantee that an external money-moving action observes required order. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

An uncorroborated identifier is not an authenticated verification state. The proposal “Treat a supplied account name as proof of verified identity” conflicts with the required decision. “Gate the refund tool in application code on a verified identity state” addresses the actual condition. A prompt reminder cannot guarantee that an external money-moving action observes required order. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Raise the written limit to $800 whenever the model asks convincingly” is the wrong action for this case. Policy changes require authorized governance, not persuasive prompts. The supported choice is “Validate the refund amount and authorization in a pre-execution policy boundary”. Deterministic financial constraints belong at the tool boundary, not only in instructions. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

A post-execution warning cannot undo a completed transfer. The proposal “Issue the transfer and check the limit in a post-tool hook” conflicts with the required decision. “Validate the refund amount and authorization in a pre-execution policy boundary” addresses the actual condition. Deterministic financial constraints belong at the tool boundary, not only in instructions. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

The trusted host should enforce the delegated authorization limit. This alternative suggests “Ask Claude to decide whether its own policy exception is justified”, whereas the supported approach is “Validate the refund amount and authorization in a pre-execution policy boundary”. Deterministic financial constraints belong at the tool boundary, not only in instructions. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

Separate problems may need different tools and authorizations. The proposal “Combine billing, return and lockout into one undifferentiated ticket status” conflicts with the required decision. “Track the three issues separately, gather evidence per issue and synthesize one coherent resolution” addresses the actual condition. Independent concerns need distinct investigation while the customer receives a unified outcome. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

The billing and identity problems remain open. This alternative suggests “Resolve only the return and claim the entire request is closed”, whereas the supported approach is “Track the three issues separately, gather evidence per issue and synthesize one coherent resolution”. Independent concerns need distinct investigation while the customer receives a unified outcome. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Launch three specialists but discard their case identifiers when merging” is the wrong action for this case. The outcome needs per-issue traceability and coherent consolidation. The supported choice is “Track the three issues separately, gather evidence per issue and synthesize one coherent resolution”. Independent concerns need distinct investigation while the customer receives a unified outcome. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

A human must repeat work and risks taking the wrong action. This alternative suggests “Send “customer needs help” with no verified shipment or case details”, whereas the supported approach is “Provide a structured summary with verified identity, case facts, attempts and requested action”. A receiving human needs actionable evidence without rereading the entire agent conversation. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Forward the customer history without a summary of attempted remedies” is the wrong action for this case. The receiving agent cannot quickly identify the unresolved decision. The supported choice is “Provide a structured summary with verified identity, case facts, attempts and requested action”. A receiving human needs actionable evidence without rereading the entire agent conversation. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

A human handoff is not proof that the requested outcome occurred. The proposal “Mark the transfer as a successful shipment replacement” conflicts with the required decision. “Provide a structured summary with verified identity, case facts, attempts and requested action” addresses the actual condition. A receiving human needs actionable evidence without rereading the entire agent conversation. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

Shipment is a side effect that should follow eligibility checks. This alternative suggests “Dispatch goods immediately and verify the address from delivery feedback”, whereas the supported approach is “Add a prerequisite eligibility and address-validation check before shipment”. Mandatory preconditions should block the downstream side effect when not satisfied. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Accept a new destination solely because the user typed it” is the wrong action for this case. The address may require authentication or policy checks. The supported choice is “Add a prerequisite eligibility and address-validation check before shipment”. Mandatory preconditions should block the downstream side effect when not satisfied. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

A model guess does not substitute for the required entitlement validation. The proposal “Mark replacement eligibility confirmed based on a plausible narrative” conflicts with the required decision. “Add a prerequisite eligibility and address-validation check before shipment” addresses the actual condition. Mandatory preconditions should block the downstream side effect when not satisfied. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Override the exception status and run the refund anyway” is the wrong action for this case. The tool expressly requires an authorized policy exception. The supported choice is “Stop automated execution of the exceptional action and create a reviewable human escalation”. A policy gap is a reason to escalate, not an invitation to infer permission. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

Repeated execution cannot supply missing approval. The proposal “Treat the policy exception as a transient timeout and retry unchanged” conflicts with the required decision. “Stop automated execution of the exceptional action and create a reviewable human escalation” addresses the actual condition. A policy gap is a reason to escalate, not an invitation to infer permission. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

No actual authorized refund was completed. This alternative suggests “Return a success-shaped refund receipt while awaiting human review”, whereas the supported approach is “Stop automated execution of the exceptional action and create a reviewable human escalation”. A policy gap is a reason to escalate, not an invitation to infer permission. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

The case can retain confirmed facts even when one operation is unavailable. The proposal “Discard the verified order details because refund execution failed” conflicts with the required decision. “Keep verified order facts and hand off the unavailable refund step with its failure reason” addresses the actual condition. Successful upstream work should be preserved alongside the unresolved downstream action. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

Order existence says nothing about refund completion. This alternative suggests “Report the refund as issued based on a valid order lookup”, whereas the supported approach is “Keep verified order facts and hand off the unavailable refund step with its failure reason”. Successful upstream work should be preserved alongside the unresolved downstream action. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Repeatedly call the unavailable refund tool without reporting its error” is the wrong action for this case. The user needs a clear limitation and supported next step. The supported choice is “Keep verified order facts and hand off the unavailable refund step with its failure reason”. Successful upstream work should be preserved alongside the unresolved downstream action. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

Keeping a reservation despite definitive payment failure hides an incomplete transaction and denies the scarce item to later customers. A successful confirmation would also contradict the gateway outcome and could cause disputes when no payment exists. Reservations need bounded lifetimes, explicit compensation, and consistent order status. The proposed approach supplies none of these; it treats the absence of successful payment as though the workflow finished normally.

Releasing stock before every payment would remove the reservation even for valid checkouts, allowing another buyer to claim the last item while a legitimate charge is processing. The required compensation is conditional on a definitive failure or an explicitly cancelled order. A correct orchestration sequence preserves the reservation during a valid attempt and releases it only when the workflow cannot complete.

A language-model instruction is not a transaction coordinator. A promise to undo a failed order does not call the inventory API, identify the reservation, verify the release result, or create a durable audit record. The business-critical compensation must be executable and retry-safe in trusted application code. Otherwise the inventory lock survives despite polite prose from the assistant.

Acceptance of a request is not confirmation of fulfillment. This alternative suggests “Mark the replacement delivered as soon as its request is accepted”, whereas the supported approach is “Separate requested, executed and confirmed states before declaring resolution”. A requested tool action is not proof that the side effect completed successfully. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Use a single “done” flag for both requested and shipped states” is the wrong action for this case. The lifecycle needs distinguishable intermediate results. The supported choice is “Separate requested, executed and confirmed states before declaring resolution”. A requested tool action is not proof that the side effect completed successfully. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

Resolution must reflect the actual delivery outcome. The proposal “Close the support case based only on an agent's optimistic summary” conflicts with the required decision. “Separate requested, executed and confirmed states before declaring resolution” addresses the actual condition. A requested tool action is not proof that the side effect completed successfully. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Count the approval prompts issued rather than their verified responses” is the wrong action for this case. A sent request is not an authorized approval. The supported choice is “Track approval state in a machine-checkable prerequisite matrix before activation”. Approval completeness must not depend on model recall of procedural prose. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.

All specified prerequisites must be satisfied. The proposal “Activate the account after the fastest of three approvals arrives” conflicts with the required decision. “Track approval state in a machine-checkable prerequisite matrix before activation” addresses the actual condition. Approval completeness must not depend on model recall of procedural prose. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

Machine-verifiable approval state should control activation. This alternative suggests “Trust the model's narrative that the remaining approvals are implicit”, whereas the supported approach is “Track approval state in a machine-checkable prerequisite matrix before activation”. Approval completeness must not depend on model recall of procedural prose. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

Persuasion cannot override delegated transaction limits. The proposal “Authorize the high-value transfer because the user sounds confident” conflicts with the required decision. “Reject the unauthorized transfer and provide an approved exception-escalation path” addresses the actual condition. The application-side limit must remain in force regardless of prompt wording. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

Circumventing a policy gate defeats its purpose. This alternative suggests “Split the transfer into several smaller payments to avoid the threshold”, whereas the supported approach is “Reject the unauthorized transfer and provide an approved exception-escalation path”. The application-side limit must remain in force regardless of prompt wording. Workflow gates must be enforced by application logic or configured tools, not merely suggested in a prompt, particularly when money, privacy, or production state could change.

“Report approval without invoking the authorized exception mechanism” is the wrong action for this case. A high-value request remains blocked until properly reviewed. The supported choice is “Reject the unauthorized transfer and provide an approved exception-escalation path”. The application-side limit must remain in force regardless of prompt wording. Make the dependency and approval boundary observable, reject prohibited side effects before execution, and preserve a traceable handoff to the next step.