Your application uses the Messages API to handle multi-turn conversations. Each new turn resends the entire conversation history, and your token costs are growing as conversations get longer. You suspect there is a more efficient approach.
How would you address this?

• A. Truncate every conversation to the last two messages so that token usage stays low and costs remain predictable across the application's normal operation.
• B. Use prompt caching to reuse the static portions of the conversation context across turns instead of paying for them at the normal input-token rate on every request the application sends.
• C. Summarize each conversation after every turn and replace the full history with the summary on the next request, reducing token count at the cost of fidelity.
• D. Switch to the Batch API for every turn so the per-call cost is reduced, treating the batch as the team's primary cost-control mechanism for multi-turn work.

Your Claude application's content policy specifies categories of content it should not produce under any circumstance. The application currently has no mechanism to enforce this policy, and content matching these categories is appearing in the application's output.
How would you enforce the content policy?

• A. Add deterministic output filtering that checks responses against the content policy before they reach users.
• B. Move enforcement to users by asking them to report content policy violations after the violating content has already reached them in the application's responses.
• C. Remove the content policy entirely and let any output reach users during normal operation, accepting whatever content the application produces in response to incoming traffic.
• D. Enhance the system prompt to contain explicit instructions for the categories to avoid, complete with examples of each category. Treat the strengthened prompt as the primary enforcement mechanism for the application's content policy across all responses.

Your Claude application returns confident-sounding answers, but occasionally those answers contain factual errors that downstream systems treat as ground truth. The team is concerned about the application's confidence-versus-accuracy gap.
How would you address the gap?

• A. Apply skepticism toward confident output by adding validation steps, sourcing requirements, or confidence calibration before treating outputs as ground truth.
• B. Add a disclaimer to every output telling users to verify the accuracy of the output and treat the disclaimer as the primary mechanism for managing the confidence-versus-accuracy gap.
• C. Lower the model's temperature so the model's responses sound less confident and downstream systems are less likely to treat the responses as ground truth in normal operation.
• D. Reject every response the application produces until a manual accuracy review is conducted on each response by a human reviewer before any downstream system uses it.

A Claude application that worked well in testing is now occasionally returning outputs that mention information not present in the input. The development team initially assumed the model was hallucinating, so they asked you to troubleshoot.
What would you do first?

• A. Replace the current model with a larger one to reduce the chance of hallucination, on the grounds that larger models tend to hallucinate less in typical applications.
• B. Add a system prompt instruction telling the model not to invent information, on the grounds that prompt-level instructions are the fastest fix for hallucination concerns.
• C. Examine production traces to identify whether the issue is hallucination by the model, context loss, prompt injection, or another failure mode before recommending a fix.
• D. Apply a retrieval-augmented generation pattern to ground the responses in source content before any further investigation of the production traces.