This might be the AI you
wanted for your tennis.
The chores. Finding a court that will actually be playable tonight. How long the light lasts. Booking your coach. And learning to read your game, your swing, the way you train.
Mental game and breathing are being built next.
Do not take her word for it.
Ask about a real court. Then paste the same question into any other assistant and see what comes back.
Pick a city and Haven answers from the index.
You may get a different answer than we did, and a different one again tomorrow. That is the finding, not a bug.
4,292 venues · 8,803 courts · 249 confirmed lit · 1,363 we have not checked, and say so
Why she can and others cannot.
She is not a bigger model. She is a record, and code that computes over it. We measured the split: on 240 questions with the right answer computed rather than judged, a frontier model handed our own data still counts courts inside a radius 2 times in 8, and works out when the light goes 1 time in 11.
Retrieval and arithmetic to code. Language to the model. That is the whole design, and it is the same finding on a tennis court as on our bench.
Check us, then use us.
The exam is public. Download it, run it on your own machine, point it at any assistant you like, and see the table for yourself.
When you want Haven doing the chores rather than answering questions, a season is $25 — your own key, your own reference, your courts and your week.
Tell us what it got wrong. That is the fastest way to make it better.