CCAO-F Practice Test 4 — 25 Questions

Question 1

A team is building a support triage assistant on the Claude API. They want the model to classify each ticket into a fixed set of categories and also return a short justification, while guaranteeing that the category value is always one of five allowed strings. They are choosing between tool use with a JSON schema and free-form text output parsed with a regex. Which TWO statements correctly describe the advantages of the tool use approach in this scenario? (Choose two.)

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

The tool input schema constrains the model's output to the declared properties and types, so the category field can be defined as an enum of the five allowed values.. Defining a tool whose input schema declares the category as an enum and a justification as a string gives the application a structured, schema-shaped payload with the allowed values baked into the contract, and it removes brittle regex scraping. The remaining claims are false: schema guidance still warrants validation, structured output is not billed differently, and natural-language text is not forcibly suppressed around a tool call.

Answer

Tool use guarantees the model will never produce an invalid category, so the application can skip validating the returned value before storing it.

Answer

Enabling tool use automatically reduces the token cost of each request because structured outputs are billed at a lower rate than free-form text.

Answer

Tool use forces the model to return only the tool call and suppresses any accompanying natural-language explanation, which is why the justification must be placed inside the tool input.

Question 2

A financial analyst uses Claude to summarize a 60-page quarterly earnings report. The prompt includes the full report between <document> tags and asks for a 200-word summary of key risks. Claude's summary frequently omits risks mentioned in the middle of the report. What is the most effective change to the prompt to improve recall of mid-document content?

Accepted Answer

Split the report into smaller sections, summarize each section separately, then combine the summaries in a final pass.. Long documents can suffer from the lost-in-the-middle effect, where content in the center receives less attention. Splitting the document into smaller sections and summarizing each ensures that every part is processed with sufficient focus. A final aggregation step combines the section summaries, preserving mid-document risks that would otherwise be missed.

Answer

Move the instruction and the risk summary request to the end of the prompt, after the document.

Answer

Lower the temperature to 0 to make Claude more deterministic and factual.

Answer

Increase the max_tokens parameter to allow a longer summary that can include more risks.

Question 3

A team is using Claude to generate SQL queries from natural-language questions against a complex schema. They want to improve correctness on multi-table joins. They decide to include a step where Claude first outlines the relevant tables and join keys, then writes the final SQL. Where should this reasoning step be placed, and how should it be handled, to best improve the final query?

Accepted Answer

Have Claude generate the reasoning outline first in a separate step, then in a second step provide that outline along with the question to produce the final SQL.. For complex generation tasks, separating planning from the final deliverable improves both quality and usability. Producing the outline in a first step and then using it in a second step to generate SQL keeps reasoning out of the executable output and lets the model focus on each stage. Combining reasoning with the final SQL, generating multiple candidates, or placing the outline after the SQL all fail to provide the same planning benefit.

Answer

Ask Claude to produce the outline and the final SQL together in one response, then have the application extract only the SQL portion.

Answer

Instruct Claude to skip the outline and instead generate three different SQL queries, then choose the one that looks most efficient.

Answer

Place the outline instruction after the final SQL in the prompt so Claude can verify its query against the outline before responding.

Question 4

A developer is using Claude to summarize legal contracts. The contracts can be very long, sometimes exceeding 100,000 tokens. The developer wants to ensure that Claude focuses on the most relevant sections, such as indemnification and termination clauses, while ignoring boilerplate. Which approach is best for managing the context window and improving summary accuracy?

Accepted Answer

Use a retrieval system to extract sections containing keywords like 'indemnification' and 'termination' and include only those in the prompt.. For very long documents, using a retrieval system to extract only the sections relevant to the summarization task is the most effective approach. It reduces the context size, ensuring Claude can process the entire input within token limits, and focuses attention on critical clauses. This method improves accuracy by filtering out boilerplate and irrelevant text, allowing Claude to generate a concise and relevant summary.

Answer

Increase the temperature to encourage Claude to explore different parts of the contract.

Answer

Ask Claude to first summarize each page of the contract and then combine the summaries.

Answer

Place the entire contract in the prompt and ask Claude to summarize the key clauses.

Question 5

A marketing team wants to use Claude to generate personalized email campaigns. They plan to include customers' full names, email addresses, and purchase histories in the prompts. Which practice best aligns with responsible data handling?

Accepted Answer

Minimize personally identifiable information (PII) in prompts by using anonymized or aggregated data, and ensure compliance with relevant privacy regulations.. Responsible data handling with Claude involves minimizing PII in prompts and complying with privacy regulations. Using anonymized or aggregated data reduces exposure while still supporting personalization. Encrypting data, instructing the model to forget, or isolating instances do not address the fundamental risk of sending unnecessary personal information. Data minimization is the key practice.

Answer

Use a separate Claude instance for each customer to isolate their data, without changing what data is included in prompts.

Answer

Send full PII but instruct Claude to forget the data after generating the email, relying on the model's ability to discard information.

Answer

Include all customer data in the prompt to maximize personalization, since Claude's API encrypts data in transit.

Question 6

A product manager is preparing to launch a Claude-powered assistant that will summarize user-uploaded medical lab reports. Before release, legal asks the team to confirm that the assistant will not provide direct diagnoses or treatment recommendations. The team decides to add a system prompt that instructs Claude to avoid giving medical advice and to recommend consulting a licensed clinician. Which action best aligns with Anthropic's safety guidance for this deployment?

Accepted Answer

Implement the system prompt, then run adversarial and edge-case evaluations to verify refusal behavior and monitor outputs after launch.. The correct approach layers a clear system prompt with empirical validation and post-deployment monitoring. Instructions alone cannot guarantee safe behavior in a high-stakes domain, so teams must test for refusal consistency and watch for regressions. This combination of preventive prompting and detective controls reflects responsible deployment practices for medical-adjacent use cases.

Answer

Rely solely on the system prompt and ship without any further testing, because Claude's Constitutional AI training already prevents medical advice.

Answer

Disable Claude's ability to discuss health topics entirely by blocking any prompt containing medical terminology before it reaches the model.

Answer

Ask Claude to role-play as a licensed physician so that its medical summaries are more authoritative and useful to end users.

Question 7

A developer is using Claude to generate synthetic data for training a fraud detection model. They want to ensure the synthetic data does not contain real personally identifiable information (PII) from the original dataset. What is the best approach?

Accepted Answer

Prompt Claude to generate data that is statistically similar but does not include any real PII, and then run a PII detection tool on the output before use.. The best approach is to instruct Claude to generate data without real PII and then validate the output with a PII detection tool. This dual strategy minimizes the chance of privacy leaks. Assuming safety, adjusting temperature, or fine-tuning on the original data all fail to provide reliable protection against PII reproduction.

Answer

Use a higher temperature setting to make the output more random and less likely to contain real PII.

Answer

Fine-tune Claude on the original dataset so it learns the patterns without memorizing PII.

Answer

Assume Claude will never reproduce PII because it is trained to avoid such outputs.

Question 8

A product team is evaluating Claude for a customer-facing assistant that must refuse to give medical diagnoses. Which TWO techniques should they use to make the refusal behavior consistent across many different user phrasings? (Choose two.)

Accepted Answer

Include a few-shot example showing the assistant declining a diagnosis request.. Consistent refusal behavior comes from persistent instructions plus concrete demonstrations. A system-prompt policy establishes the rule for every turn, while a few-shot example shows exactly how a refusal should look, helping the model generalize across varied user phrasings. Temperature, max_tokens, and removing the system prompt do not reinforce policy adherence and can weaken it.

Answer

Set temperature to 1.0 to encourage diverse safe responses.

Answer

Increase max_tokens so the model has room to explain refusals.

Answer

Disable the system prompt and rely only on user instructions.

Question 9

A logistics company uses Claude to extract shipment details from scanned customs forms. The forms are supplied as raw OCR text that contains occasional garbled characters and spurious line breaks. The developer wants to reduce the number of fields Claude invents when a value is missing on the form. Which prompt structure change is most likely to achieve this?

Accepted Answer

Wrap the OCR text in <document> tags and add the instruction: 'If a field cannot be found in the document, return null for that field.'. The reliable fix pairs two techniques: structural delimitation of the source text so Claude can tell document content from instructions, and an explicit rule stating what to do when a field is absent. Together they remove the ambiguity that pushes the model toward inventing values, and the null convention makes the result easy to validate programmatically.

Answer

Shorten the OCR text by removing all line breaks before inserting it into the prompt.

Answer

Set temperature to 0 and request the output as a JSON object.

Answer

Add the instruction: 'Be as accurate as possible and avoid mistakes.'

Question 10

An engineer is designing a multi-tenant SaaS feature that lets each customer supply their own Anthropic API key, which the backend stores encrypted and uses when calling the Messages API on that tenant's behalf. A security review asks how requests should be attributed so usage can be billed back and abuse isolated per tenant. Which practice best meets this requirement?

Accepted Answer

Have the backend call the Messages API with each tenant's own stored key so rate limits, usage, and audit trails are scoped to that tenant.. Credential boundaries are the cleanest way to achieve per-tenant attribution and isolation. When each tenant's traffic flows through that tenant's own API key, Anthropic's rate limiting, usage records, and audit logs partition automatically, and a compromised or abusive tenant cannot exhaust capacity belonging to others. The backend retains control over request construction and can layer additional quotas.

Answer

Proxy all tenant traffic through a single key and reconcile billing by counting tokens in the responses your backend receives.

Answer

Issue each tenant the same platform key but append a unique tenant ID to the user field of every message.

Answer

Send every tenant's request using a single shared platform API key and rely on the prompt contents to identify the tenant.

Question 11

A team is deploying Claude to summarize internal incident reports that may contain sensitive employee information and security vulnerabilities. Which TWO practices best align with responsible use of Claude in this context? (Choose two.)

Accepted Answer

Redact or anonymize sensitive employee and security details before sending content to the Claude API.. Responsible use when handling sensitive incident reports involves minimizing data exposure and adding safeguards. Redacting sensitive details before sending to the API reduces risk, and output filtering catches any sensitive information that may still appear in summaries. Instructing the model to ignore data, storing logs insecurely, or sharing raw data externally do not adequately protect sensitive information and may introduce compliance issues.

Answer

Enable logging of all prompts and responses for auditing, and store them indefinitely in an unencrypted database.

Answer

Share the raw incident reports with a third-party analytics service to cross-check Claude's summaries for accuracy.

Answer

Use a system prompt to instruct Claude to ignore any sensitive information it encounters and not include it in summaries.

Question 12

An analyst is using Claude to answer questions over a 50,000-token legal contract. They notice that answers about clauses near the middle of the document are less accurate than those about the beginning or end. Which strategy best improves accuracy across the entire document?

Accepted Answer

Break the contract into labeled sections and retrieve only the relevant sections for each question.. Long documents can suffer from uneven attention, with mid-document content sometimes under-weighted. Breaking the contract into labeled sections and retrieving only relevant passages for each question ensures the needed clause is prominent in the prompt. This retrieval-based structuring improves accuracy regardless of where the clause originally appeared.

Answer

Increase the model's temperature slightly to broaden its attention.

Answer

Repeat the question at the beginning and end of the prompt around the contract.

Answer

Summarize the contract first and then ask questions only against the summary.

Question 13

A team is designing a Claude prompt that must return a fixed JSON schema with fields "vendor", "amount", and "currency". They want the output to be reliably parseable by downstream code. Which TWO techniques best improve reliability of the structured output? (Choose two.)

Accepted Answer

Use a prefilled assistant turn that begins the JSON, so Claude continues from a known starting point.. Reliable structured output comes from constraining the response shape on two fronts: telling Claude exactly what schema to produce, and preventing it from starting with prose. An explicit schema or example JSON defines the contract, while prefilling the assistant turn with the opening of the JSON forces the model to continue in that format. Together they minimize drift and make downstream parsing dependable, unlike loosening sampling or inviting commentary.

Answer

Set temperature to its maximum so Claude considers many possible JSON layouts.

Answer

Precede the JSON with a short natural-language explanation so a human can verify the result.

Answer

Ask Claude to decide at runtime which fields are most relevant and include only those.

Question 14

A support team wants Claude to answer customer questions using only the company's internal help articles. They plan to paste the relevant article text into the prompt before the user's question. Which prompting technique does this describe?

Accepted Answer

Retrieval-augmented generation (RAG). Inserting relevant source documents into the prompt so the model answers from them is retrieval-augmented generation. The retrieval step supplies factual context, and the generation step produces an answer grounded in that context. This differs from zero-shot, few-shot, and chain-of-thought, which address examples, demonstrations, and reasoning traces respectively.

Answer

Zero-shot prompting

Answer

Chain-of-thought prompting

Answer

Few-shot prompting

Question 15

A developer is designing a tool use workflow with the Claude Messages API. After the model returns a tool_use content block, the application executes the tool and must send the result back. Which TWO of the following are required for the follow-up request to be processed correctly? (Choose two.)

Accepted Answer

Preserve the prior assistant turn, including the tool_use block, in the messages array of the follow-up request.. Completing a tool use loop requires sending the tool's output back as a tool_result content block inside a user message, with a tool_use_id that matches the model's tool_use block id, while keeping the original assistant turn in the messages array. This pairing lets the model associate the result with its earlier call and continue reasoning.

Answer

Re-declare the full tools array in the follow-up request even though tools were already provided in the first request.

Answer

Set the stop_sequences parameter to include the tool name so the model knows when to stop calling tools.

Answer

Convert the tool_result content into plain assistant text so the model can read it as part of its own previous reply.

Question 16

A developer is building a customer support assistant using Claude. The assistant must always respond in a friendly tone, never discuss competitors, and always ask for an order number when the user reports a shipping issue. The developer wants these rules to apply across all conversations with minimal per-request token cost. What is the most appropriate mechanism?

Accepted Answer

Place the rules in the system prompt so they are applied consistently to every request.. The system prompt is the correct place for persistent behavioral instructions that should apply to every request. It is processed by the model as a high-level directive and is not repeated in each user message, making it token-efficient. This ensures consistent tone, competitor restrictions, and required questions across all conversations without per-request repetition.

Answer

Add the rules to the end of each user message as a reminder.

Answer

Include the rules as a few-shot example in every user message.

Answer

Fine-tune a custom model with the rules embedded in training data.

Question 17

A developer is integrating Claude into a legal document review system. The system must process lengthy contracts and answer questions about specific clauses. Which TWO of the following techniques help mitigate the risk of Claude hallucinating details not present in the document? (Choose two.)

Accepted Answer

Instruct Claude to answer only based on the provided document and to say 'I don't know' if the information is not present.. To reduce hallucinations when reviewing legal documents, the most effective methods are to instruct Claude to rely solely on the provided text and to require citations for its answers. These techniques ground the model's responses in the source material, making it less likely to invent details and easier to verify accuracy.

Answer

Use a higher temperature setting to encourage more creative responses.

Answer

Set max_tokens to a very low value to limit the response length.

Answer

Fine-tune the model on a dataset of legal contracts.

Question 18

A developer is integrating the Claude API into a production pipeline and needs to handle a response where the model stopped because it hit the output token ceiling before finishing its answer. Which TWO actions are appropriate? (Choose two.)

Accepted Answer

Inspect the 'stop_reason' field and treat a value indicating the token limit as a truncation signal.. When generation stops because it reached the output token ceiling, the response still returns successfully but the content is incomplete. The application should detect this via stop_reason and then either allow more output tokens or continue from the partial text. Retrying unchanged, adjusting temperature, or trusting the HTTP status all fail to detect or resolve the truncation.

Answer

Assume the response is complete and parse it as-is, since a 200 status was returned.

Answer

Retry the identical request unchanged, expecting the model to finish within the same limit.

Answer

Lower 'temperature' to 0 so the model produces a shorter answer next time.

Question 19

A developer is building a Claude-powered assistant for a legal firm. The assistant is asked to draft a clause citing a specific statute. Claude generates a citation that appears authoritative but does not actually exist. The developer wants to reduce the risk of this happening in production. Which approach is most aligned with responsible use?

Accepted Answer

Use retrieval-augmented generation (RAG) to ground Claude's responses in a verified legal database, and add automated checks that validate citations against that database.. Hallucinated citations in legal drafting pose serious risks. Retrieval-augmented generation grounds Claude's responses in a verified database, and automated checks validate that cited statutes exist. This combination addresses the root cause by ensuring outputs are based on real sources and are programmatically verified. Disclaimers, higher temperature, or limited fine-tuning do not reliably prevent fabricated references in production.

Answer

Instruct Claude to always add a disclaimer that its output may contain errors, and rely on the attorney to verify every citation manually.

Answer

Increase the model's temperature setting so that Claude produces more creative and varied citations, reducing the chance of repeating a fake one.

Answer

Fine-tune Claude on a small set of real legal documents and assume it will generalize to all statutes and jurisdictions without further validation.

Question 20

A developer is implementing retry logic for calls to the Anthropic Messages API in a production service. The service must handle transient failures gracefully without overwhelming the API. Which TWO practices should the developer implement? (Choose two.)

Accepted Answer

Retry requests that fail with HTTP 429 and HTTP 500 status codes, using exponential backoff with jitter between attempts.. Resilient retry logic targets transient failures such as 429 and 500 responses, spacing attempts with exponential backoff plus jitter to avoid synchronized retry storms. A retry cap, whether by count or total time budget, prevents runaway loops and bounds latency. Client errors like 400 and 401 are not transient and should not be retried, and the API does not redeliver failed requests, so some client-side retry strategy is necessary.

Answer

Disable all retries and rely on the API's internal queueing to eventually deliver failed requests.

Answer

Retry all failed requests immediately in a tight loop until they succeed, to minimize latency for the end user.

Answer

Retry requests that fail with HTTP 400 and HTTP 401 status codes, since these often resolve on a second attempt.