Prompting and evals · L2 · 20 min
A golden set
Learn in five minutes
- Ten to fifty checks beat one demo.
- Each item has an input, an expected property, and a source.
- The property can be 'refuses' or 'cites this note'.
- Lab 03 is a local example with a fixed file.
- Do not add items after you see the score and pretend the set was always that.
Write the set before you tune. If you add a case because the model failed it, mark that case as new so the old score stays comparable.
Visual
Five lines on paper
- Write five questions about a note you can open.
- Mark the expected answer or the expected refusal.
- Run them. Record misses.
- Change one instruction. Run the same five.
Quiz
Ecosystem
- Regression
- A check that used to pass and now fails.
Sisters: JSON schema, Rubrics, Regression tests.
The same job-shape-test card works for chat, voice, and a tool result.
The capstone reuses this card.
Config
- Write the goal in one sentence.
- Name the input and the allowed output.
- Save the trace next to the result.
Troubleshooting
- The answer looks fluent and is wrong. Check whether a source passage actually contains the claim.
- The loop never stops. Set a step cap and stop when the tool result repeats.
Sources and go deeper
- Python 3 tutorial vendor last checked 2026-10
- NIST AI Risk Management Framework spec last checked 2026-10
- Large language model — Wikipedia wikipedia last checked 2026-10