Prompt Lab

Write the tutor's prompt. See how it does with eight learners.

The moment

A learner wants the chance that at least one of several AI copies fixes a bug (each copy: 15.9%). They added the chances, so the app now says 111.3% at 7 copies. Your tutor should help them see why adding fails, without giving the answer away.

How it works

You write the tutor's prompt.
gpt-5.1 uses it to answer 8 different learners.
Claude compares each answer with our real tutor's.
You get a score out of 8.

Eight learners

The learner says one of these next. Your tutor answers each:

Scoring

For each learner, Claude compares your reply with our real tutor's reply, twice, with the order swapped. Win both: 1 point. Split: ½. It's a high bar: a generic "be encouraging and explain clearly" prompt scores 0.

Our real tutor's prompt

This is the exact prompt our server sends gpt-5.1 for this moment, about 49,000 characters. Can you beat it?

The judge looks for a reply that answers what the learner said, asks one clear question, and doesn't hand over the answer.

Show the judge's full prompt

System prompt (Claude Sonnet, answer forced to A or B):

Message, sent twice per learner with A and B swapped:

Submit

Leaderboard

#	Team	Best score (of 8)	Runs
Loading…

Talk about it

Which learner was hardest? What did your prompt say about them?
Did your score go up because the tutoring got better, or because you learned what the judge likes?
Write one reply that would win with this judge but that you'd never want a learner to get.