quiz · 4 min Prompting & Control Check Why is the instruction considered an integral part of the policy?It is decoration that makes prompts more politeChanging it changes the policy as surely as swapping model weightsIt only affects the model's tone, not its output shapeIt is cached separately and has no effect on behaviorWhat makes a parser 'total' in the context of structured output?It only produces a defined action when the model returns perfectly formatted outputIt handles every possible response, including malformed ones, mapping each to a known safe actionIt rejects any response that does not exactly match the expected format and raises an exceptionIt strips all whitespace before returning the raw model stringIn the KIND: argument pattern from Lesson 7, what happens when the parsed kind is not in the VALID set?The loop retries the prompt automaticallyThe model is called again with a stricter instructionThe response is returned as-is to the downstream branchThe parser returns ('ANSWER', 'unparseable') as a safe fallbackWhat is the best defense against output drift when a model provider updates weights?Add more adjectives like 'concise' and 'always use the exact format' to the promptA total parser that normalizes case and whitespace, plus a regression test suitePin to an older model version forever and never upgradeIncrease the prompt length to cover more edge casesWhich defensive habit does Lesson 8 recommend for untrusted external content in an instruction?Summarize it with a second model call before interpolatingLimit it to 100 characters so it cannot carry malicious instructionsUse structural separators (XML tags, delimiters, role boundaries) to signal where instructions end and data beginsLog it after the fact so you can review it if something goes wrongSubmit Mark complete ← Prompting Pitfalls What Is a Tool? →