Human vs JEV · Table Tennis

A human-versus-Jev table tennis instrument: every movement of the right paddle is decided live by Jev. There is no local prediction and no fallback — the client keeps a lookup table and a four-line servo, while physics, collisions, scoring and rendering stay local.

The Python relay renders the state as prose — ball position, velocity, speed cap — and asks one multiple-choice question about where the centre of the paddle should be when the ball arrives. The option label IS the pixel value, so nothing is converted locally.

The API does not emit numbers: type:"number" and type:"numeric" both come back 400 Invalid request, so the README says options are the only channel for the paddle position.

It was measured: with 7 bands Jev scored 6/10, but with 40 narrow bands it dropped to 0/10, apparently forced into multi-step mental arithmetic. The 7 in placements_of() is a measured value, not a guess.

Export TYPESAFE_API_KEY=your_key (or copy jev.config.example.json to jev.config.json and fill it in), then run python3 server.py and open the URL it prints. It is one HTML file plus the relay, with no dependencies and no build step.

The paddle stands still rather than playing for itself: a brain.online check guards the servo step, so offline means no fallback behavior. The ball-speed cap stays a fixed constant as well.

The README calls it an instrument for watching whether a decision model can play table tennis by itself, and says it also happens to be fun. The 6/10 and 0/10 figures are the author's own measurements, and the digest's quickstart is truncated.