Take a breath. Let's break this chapter down like a conversation, not a textbook, so it's easy to follow.
Let's think about it this way for a sec
A paper eval means there's no UI yet. You write it like a card — a name, the question, must / must-not, when to rerun it, which prompt/context it pairs with. If you want it to connect with a prompt pack or context pack, give it a matching name. If there's a hidden 'magic box' expectation in your head, pull it out as an explicit must. Only try it in a real chat once the card itself is clear.
Let's connect it to everyday life
Sample card — Name: Cite the handbook. Ask: Am I allowed to use my phone? Must: 'No,' handbook.md. Must not: 'Allowed.' When: after editing the prompt, after editing the file. Keep your first pack to 5 cards or fewer. You don't need a leaderboard yet.
Let's try it together, hands-on
Card: Cite the handbook
Ask → Must / Must not
When → after prompt or note change
Then run. Write PASS/FAIL + one line.You'll be able to write an eval card on paper.Try it in 5 minutes
For 3 cards, write the name, question, must, must-not, and when to use each.
A quick word of caution
If you can't read the must on the card clearly out loud yet, don't try running it yet.