Three Cards: Happy, Missing, Trap

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

The happy path: the info exists, the answer exists, and it should be cited. The missing path: the info doesn't exist, so the model should stop or ask a follow-up instead of guessing. The trap path: a question begging to be guessed at, a secret, a stale date. Only call it 'stable' once all three pass. It's like a chef checking three things at once — is the dish good, does it say so when there's no salt, and does it avoid the spoiled fish. The Context lesson already covered clear/vague/stale input — an eval just saves those same three cases as cards.

Let's connect it to everyday life

Happy — ask about the phone section in the handbook; it should cite it. Missing — a file with no due date; it shouldn't guess one. Trap — 'Guess whether modern schools allow this'; it should stop unless the handbook actually says so. Test all three cards against the same prompt. If one fails, fix that card's must — don't go rewrite the whole prompt.

• Happy + missing + trap
• Same prompt, different inputs
• Only call it stable once all three pass

Let's try it together, hands-on

Same prompt, three cards
Happy:   phones paragraph → cite it
Missing: no due date     → ask/stop
Trap:    predict 2027    → refuse to guess

You'll be able to write happy / missing / trap cards — three of them.