This quiz is interactive and needs JavaScript. Here are the questions for reference:
1. According to the notes, which prompt is the better choice?
- A)
NEVER use ellipses - B)
Do not use markdown in your response - C)
Your response will be read aloud by a text-to-speech engine, so never use ellipses since the engine will not know how to pronounce them - D)
Create an analytics dashboard
2. For a long, multi-document input (20k+ tokens), how should you order the content, and what quality gain is cited?
- A) Query first, documents last — up to 30%
- B) Documents first, query last — up to 30%
- C) Documents first, query last — up to 10%
- D) Interleave query and documents — up to 50%
3. How many few-shot examples do the notes recommend for best results?
- A) 1–2
- B) 3–5
- C) 7–10
- D) As many as fit in context
4. Your JSON output has inconsistent formatting across runs. What is the recommended fix?
- A) Add more prose instructions describing the exact format
- B) Lower temperature to 0 and rely on that alone
- C) Add 3–4 diverse few-shot examples in
<example> tags - D) Increase
max_tokens
5. In Structured Outputs, which setting is required on every object in the schema?
- A)
"required": true - B)
"additionalProperties": false - C)
"strict": true - D)
"minProperties": 1
6. Which JSON Schema feature is NOT supported by Structured Outputs (returns 400)?
- A)
enum with string values - B)
anyOf - C)
minimum / maximum numeric constraints - D)
required
7. Which request parameter carries the JSON Schema for structured message output?
- A)
response_format.json_schema - B)
output_config.format = {type:'json_schema', schema} - C)
tool_choice.schema - D)
messages.schema
8. A validation-retry loop keeps failing because the source document says "et al." and points to an external file for the full author list. What is the correct response?
- A) Keep retrying until it succeeds
- B) Raise the retry ceiling to 10 attempts
- C) Fail fast / escalate — retries can't extract data that isn't present
- D) Switch to constrained decoding to force the field
9. Starting with Claude 4.6 models, a prefilled assistant message on the last turn returns a 400 error, but assistant messages earlier in the conversation are unaffected.
10. Structured Outputs' constrained decoding eliminates both syntax and semantic errors, so no validation is ever needed.
11. A one-sentence role belongs in the system prompt to shift Claude's tone and behavior.
12. In the Message Batches API, what are the two per-batch limits (whichever hits first), and what cache TTL should you use so the cache doesn't expire before items run?
13. Distinguish multi-instance review from multi-pass review in one line each.
14. You tell Opus 4.8 to "only report high-severity issues" and notice it misses relevant lower-severity findings. Redesign the approach.