Unit ID: PAW-M02-U04 Estimated active time: 20-25 minutes

Good criteria are not enough

Success criteria describe what good output should do.

Failure criteria describe what would make output unusable.

Both are useful.

When you define failure criteria before generation, you are less likely to accept a polished but risky answer.

What are failure criteria?

Failure criteria are conditions that make the output weak, unsafe, or not usable.

Examples:

• It invents facts.
• It includes private information.
• It promises something not approved.
• It uses the wrong tone.
• It is too vague to act on.
• It ignores the audience.
• It gives a final recommendation when facts are missing.
• It hides assumptions.
• It gives advice that needs a qualified expert.

Failure criteria help you reject output

Many AI answers sound confident.

If you do not define failure criteria, you may accept output because it feels polished.

Failure criteria give you permission to say:

> This sounds good, but it is not usable.

That is a strong review habit.

Example

Task:

> Create a simple comparison table for three fictional workshop ideas.

Success criteria:

• includes three ideas;
• compares learner value, effort, risk, and no-paid-tool requirement;
• uses simple language;
• lists assumptions separately.

Failure criteria:

• invents real costs or dates;
• claims one idea is proven best without evidence;
• ignores risk;
• suggests collecting private learner data;
• gives a final decision without showing tradeoffs.

Now you know what to reject.

Stop conditions

A stop condition is a reason to stop using AI or ask for human review.

Examples:

• The task includes confidential information.
• The output affects a real person's job, money, health, education, safety, or legal position.
• The output includes facts you cannot verify.
• You cannot explain the answer yourself.
• The answer requires professional judgement.
• The tool asks for information you should not provide.

Stop conditions protect people and reduce careless use.

Prompt with failure criteria

You can include failure criteria in a prompt:

> If you are missing important information, do not guess. List questions instead. Do not invent dates, costs, names, evidence, or approvals. Do not give a final recommendation if the assumptions are weak.

This does not make the output perfect. But it tells the tool what not to do.

Mini practice

Choose one task from the previous unit.

Write:

• three failure criteria;
• two stop conditions;
• one sentence you can add to the prompt to reduce guessing.

Example sentence:

> If information is missing, list questions instead of inventing details.

Mark complete