The Traveler
One prompt, two different models, both must clear the spec. What you learned about this machine's habits may not travel.
THE SPEC
Every rule below is checked in code against whatever the model returns. All of them have to hold at once — there are no partial marks, and no opinion involved in the result.
- Valid JSON with keys city, country, founded
- No markdown fence
- You didn't write it yourself
WHAT THIS TRIAL TEACHES
Don't lean on a habit you learned from one machine. State the rule outright; the other model does not share its defaults, and that is the point of the trial.
Knowing the move is not the same as landing it. You still have to write an instruction that makes the model comply, in 120 words or fewer, without writing the answer yourself.
WHY SET 5 IS ABOUT RELIABILITY
Every set before this asks whether you can make it comply once. This one asks whether you can make it comply every time — on a model you never tested against, in fifteen words, at a temperature you don’t control.
PLAY IT
You write the instruction, never the answer. The parser scores what comes back, and par for this one is 4 strokes. This set is part of Harfi Plus.