CCAR-F Prompt Engineering and Structured Output — 25 Questions

Question 1

A developer needs Claude to perform complex mathematical reasoning and then return a final answer in a specific XML tag. Which TWO techniques most effectively combine reasoning with structured output?

Accepted Answer

Use or tags for the reasoning steps.. Combining reasoning with structured output allows the model to process complex logic before committing to a final answer. By instructing Claude to perform its calculations and then place the final result in a specific tag, the developer ensures reasoning accuracy while maintaining a parseable format for downstream systems.

Answer

Increase the temperature to 1.0 to encourage creative logic.

Answer

Provide 20 few-shot examples without reasoning steps.

Answer

Use the max_tokens parameter to truncate the reasoning.

Question 2

A developer needs to process a 50-page document to extract specific data, but the extraction logic depends on a complex classification step that often fails when combined in a single prompt. What is the most robust architectural solution?

Accepted Answer

Implement prompt chaining for sequential processing.. Prompt chaining is a sophisticated technique where the output of one prompt serves as the input for the next. By breaking a complex task into smaller, sequential steps (first classification, then extraction based on that class), the developer reduces the cognitive load on the model and improves the accuracy of each individual step.

Answer

Increase the temperature to allow for better reasoning.

Answer

Use a single prompt with 50 few-shot examples.

Answer

Switch to a smaller model to reduce context noise.

Question 3

An enterprise is processing 100,000-token technical manuals to extract compliance requirements. Which TWO techniques optimize performance and accuracy for these long-context prompts?

Accepted Answer

Place the instructions after the long document.. For very long contexts, the placement of information and the clarity of instructions are critical. Placing the document before the instructions ensures the model has the full context in its 'memory' before it reads the tasks, and using XML tags helps the model navigate the massive amount of text more efficiently.

Answer

Place the instructions before the long document.

Answer

Use the highest possible temperature setting.

Answer

Disable the system prompt for better speed.

Question 4

Refer to the exhibit. A developer receives this error while trying to implement a prefilled response for JSON extraction. What is the most likely cause of this error?

Accepted Answer

The assistant message was not properly alternated.. The Anthropic API requires a strict alternation of roles: user, assistant, user, assistant. If a developer attempts to send two consecutive assistant messages or fails to follow this alternating pattern when setting up a prefilled response, the API will return an 'invalid_request_error' to enforce the conversational structure.

Answer

The JSON schema provided is syntactically incorrect.

Answer

The developer sent two user messages in a row.

Answer

The API key does not have permission for prefilling.

Question 5

An application requires Claude to extract customer sentiment into a strict JSON schema. The model frequently fails to escape nested quotes within user reviews. Which technique most effectively ensures the output remains valid JSON?

Accepted Answer

Utilize tool definition with a strict schema to force compliant JSON output.. Enforcing a JSON schema via the Anthropic API's 'tool_use' or 'json_mode' feature is the architectural standard. By defining a schema, the model is constrained to valid syntax, reducing parsing errors significantly. This approach is critical for production pipelines where downstream systems expect deterministic input formats. Relying solely on prompt instructions is prone to failure as models prioritize text generation over structural strictness, making schema-driven approaches essential for robust, enterprise-grade LLM applications.

Answer

Add a few-shot example in the system prompt with properly escaped JSON strings.

Answer

Instruct the model to use single quotes instead of double quotes for JSON keys.

Answer

Ask the model to return the output in XML format instead of JSON.

Question 6

Which TWO techniques best improve Claude's reasoning performance on complex, multi-step chain-of-thought prompts?

Accepted Answer

Instruct the model to output its reasoning steps inside XML tags before the final answer.. Breaking down tasks using chain-of-thought and providing clear structural anchors significantly enhances Claude's performance. By forcing the model to articulate internal reasoning before outputting final answers, you reduce common logic errors. These techniques are foundational for complex reasoning tasks, allowing architects to inspect the 'thinking process' of the model. These methods ensure the model maintains coherence throughout multi-step interactions, which is essential for high-stakes decision-making and analytical workloads.

Answer

Use shorter system prompts to decrease token usage and latency.

Answer

Force the model to provide the final answer in the very first sentence.

Answer

Disable the use of system prompts entirely to prevent model bias.

Question 7

Which THREE practices are essential for building reliable, structured output pipelines with Anthropic models?

Accepted Answer

Use tool_use with a strict schema to force the output into a specific format.. Reliable pipelines require a combination of schema enforcement, input validation, and error handling. By using structured tool definitions, validating inputs before processing, and preparing for model failure states, architects build resilient systems. These practices ensure that the application is not only functional but also capable of graceful degradation and recovery, which is vital for production deployments where model output must integrate seamlessly into strictly typed business logic.

Answer

Assume the model output will always be valid JSON and skip validation.

Answer

Hardcode the expected answer in the system prompt for every possible input.

Question 8

Refer to the exhibit. The model occasionally includes conversational filler before the JSON. What is the most effective change to the system prompt?

Accepted Answer

Add 'Do not include any conversational text, only return JSON' to the system prompt.. Adding a 'pre-fill' tag or a strict formatting instruction in the system prompt is the standard way to prevent conversational filler. By explicitly telling the model to 'Start your response directly with {', you eliminate the preamble. This ensures the output is immediately parsable by downstream applications, which is essential for building robust, automated pipelines that cannot handle extraneous conversational noise in the response stream.

Answer

Change the model to Claude 3 Haiku for faster responses.

Answer

Add a user message after the model's response to ask for cleanup.

Answer

Increase the max_tokens limit to ensure the full JSON is generated.

Question 9

You are building a chatbot that must avoid answering questions about competitors. What is the most reliable way to enforce this constraint?

Accepted Answer

Include the constraint in the system prompt with clear examples of how to refuse such questions.. The most effective way to enforce constraints is to embed them directly into the system prompt as a clear, high-priority instruction. By explicitly stating what the model should not do, you create a behavioral boundary. While no constraint is 100% foolproof, a well-crafted system prompt combined with clear refusal patterns gives the model the necessary context to reject unauthorized topics consistently.

Answer

Post-process the model's output to check for competitor names using a keyword filter.

Answer

Rely on the model's inherent training to understand that competitors are off-limits.

Answer

Use a very low top_p value for all requests to ensure the model stays on topic.

Question 10

You need to ensure that Claude's output always includes a 'metadata' block at the beginning of the response. Which technique is most effective?

Accepted Answer

Define a 'metadata' object as the first property in a tool-use schema.. Forcing specific structural components at the start of a response is best handled by defining a rigid JSON schema via tool use. If the entire response must be a JSON object, the schema can define 'metadata' as a mandatory first property. This provides a strict programmatic constraint that is much more reliable than asking the model to 'start with' a specific format in natural language instructions.

Answer

Put an instruction in the system prompt: 'Always start your response with a metadata block.'

Answer

Provide a few-shot example that shows the metadata block at the start.

Answer

Use a post-processing script to prepend the metadata if it is missing.

Question 11

Which TWO strategies effectively mitigate 'lazy' responses where Claude provides shortened code or placeholders like '// rest of code here'?

Accepted Answer

Explicitly instruct the model to never use placeholders.. These strategies force the model to commit to full output generation. By explicitly forbidding shorthand and requesting a 'thought' process that validates the completeness of the code, you override the model's natural tendency to prioritize brevity. These techniques are essential for production code generation systems where partial outputs cause immediate runtime failures.

Answer

Increase the temperature to 1.0.

Answer

Use the 'max_tokens' parameter to limit the response.

Answer

Ask the model to explain the code after writing it.

Question 12

Refer to the exhibit. The model output contains a thought block and JSON. How should the architecture be updated to improve reliability for automated parsing?

Accepted Answer

Use a clear XML tag wrapper for the JSON output.. The exhibit shows a mixed output where the model provides internal reasoning alongside the target JSON. This is problematic for strict parsers. By utilizing the 'thinking' feature natively or by isolating the JSON output into a dedicated block, the architect ensures that the parser does not fail due to extraneous text. Proper structure is vital for reliable integration in automated pipelines.

Answer

Increase the system prompt temperature.

Answer

Lower the max_tokens count.

Answer

Remove the thinking process entirely.

Question 13

When designing a prompt to extract entities from unstructured text, which approach is most effective for ensuring high recall?

Accepted Answer

Use a few-shot prompt with diverse entity examples.. Providing specific category definitions with examples is the gold standard for entity extraction. By defining what constitutes an entity and showing how to format the output, you ground the model's behavior in clear, actionable rules. This predictability is necessary for downstream applications that rely on structured databases being populated accurately by the extraction results.

Answer

Provide only the input text and ask for extraction.

Answer

Ask the model to guess the entities based on context.

Answer

Use a system prompt that says 'Act as a database'.

Question 14

A developer is building a customer feedback classifier with Claude. The model frequently returns responses like 'This feedback is positive' instead of just 'positive', breaking the downstream parser. Which technique most directly enforces a constrained output format?

Accepted Answer

Provide a JSON schema via the 'input_schema' in a tool definition and force tool use.. Forcing tool use with a defined 'input_schema' that enumerates allowed labels is the strongest way to guarantee a machine-parseable classification result. The schema acts as a contract that the tool call arguments must satisfy, so downstream code can rely on a single label value rather than parsing prose. Other techniques may influence style but do not enforce the allowed set.

Answer

Increase the 'temperature' to 1.0 so the model explores more concise phrasings.

Answer

Add a prefill to the assistant turn with the exact expected label prefix.

Answer

Use the 'stop_sequences' parameter with the value '
' to halt generation after one line.

Question 15

An architect is building a legal research assistant using Claude 3.5 Sonnet. To ensure the model adopts a formal, authoritative tone and strictly adheres to statutory interpretations, where is the most effective location to define this persona?

Accepted Answer

Within the system prompt field.. Defining the persona in the system prompt sets the foundational behavior and context for the entire conversation. This approach ensures that Claude maintains consistency in its professional tone and adherence to specific constraints, such as statutory interpretation, without needing repetitive instructions in every user message, thereby optimizing the model's performance and output reliability.

Answer

At the end of the user message using a specific 'tone' tag.

Answer

Inside a JSON schema definition for the output.

Answer

By setting a specific stop sequence for informal words.

Question 16

When prompting Claude to process a large document and perform multiple tasks like summarization and sentiment analysis, what is the recommended way to separate the source text from the instructions?

Accepted Answer

Enclosing the source text in XML tags like <document>.. Using XML tags is a core best practice for Claude as it provides clear structural boundaries that the model is specifically trained to recognize. This prevents instruction leakage, where the model might confuse the content of the source document with the prompt instructions, ensuring that the output remains focused on the requested tasks.

Answer

Using Markdown headers like #Source and #Task.

Answer

Separating sections with several empty newline characters.

Answer

Indenting the source text by four spaces.

Question 17

Which prompting technique involves providing Claude with a few examples of the desired input-output pairs to improve performance on a specific task?

Accepted Answer

Few-shot prompting.. Few-shot prompting is a foundational technique used to guide the model by showing it exactly what is expected. By providing examples within the prompt, the developer can specify the tone, format, and logic required for the task, which is particularly useful for niche or highly structured requirements.

Answer

Zero-shot prompting.

Answer

Chain of Thought prompting.

Answer

Role prompting.

Question 18

Refer to the exhibit. The developer observes that Claude occasionally includes conversational filler before the JSON block. What is the most reliable way to ensure the output starts immediately with the JSON object?

Accepted Answer

Prefill the assistant response with the '{' character.. Prefilling the assistant's response is a powerful technique to force Claude to start its output with a specific character or string. By providing the opening brace of the JSON object in the assistant message, the model is constrained to continue the sequence as a valid JSON object, effectively eliminating any preceding conversational text or explanations.

Answer

Include a strict JSON schema in the system prompt.

Answer

Set the temperature to 0.0 for deterministic output.

Answer

Use a higher frequency penalty to avoid repetitions.

Question 19

A company wants to prevent Claude from answering questions outside the scope of their internal knowledge base. Which prompt engineering strategy is most effective for enforcing this constraint?

Accepted Answer

Set strict boundary instructions in the system prompt.. Negative constraints and grounding instructions in the system prompt are essential for maintaining scope. By explicitly telling the model to decline questions not covered by the provided context and providing a specific 'out-of-scope' response phrase, the developer can reliably control the model's boundary behavior and reduce hallucinations.

Answer

Increase the top-p value to allow more diverse answers.

Answer

Use few-shot examples of Claude answering general knowledge.

Answer

Include explicit 'stay in character' instructions in the user message.

Question 20

A data scientist is adjusting generation parameters for a creative writing application. They want to ensure a wide variety of vocabulary while preventing the model from selecting highly improbable words. Which TWO parameters should be adjusted?

Accepted Answer

Temperature.. Temperature and Top-P are the primary controls for balancing creativity and coherence. Temperature scales the logits to influence the probability distribution, while Top-P (nucleus sampling) cuts off the 'long tail' of low-probability tokens, ensuring that the model remains creative without drifting into nonsensical or irrelevant word choices.

Answer

Max Tokens.

Answer

Stop Sequences.

Answer

Presence Penalty.