Structured Extraction
Turn a messy customer message into strict JSON without inventing a value you weren't given.
Practice ground for prompt writers
Write one instruction. Run it against cases you cannot see. Then find out what broke, and what your fix broke next.
The arena
Tap a case to see what it caught.
Amount kept: The reply picked the delivery fee instead of the refund amount.
Demo board. Real runs use hidden cases and a live model.
Each one trains a different reflex. Start with the flagship.
Turn a messy customer message into strict JSON without inventing a value you weren't given.
Handle real customer requests while obeying policy, including the ones you must refuse or escalate.
Make text shorter or simpler without dropping a single fact that matters.
Three moves. About a minute per attempt.
A short instruction is enough to start.
Hidden inputs, live model, no peeking.
Watch what your edit fixes and breaks.
The first run takes about a minute. Your prompt stays in the tab, and every score comes from the model.