Sink the ball with a single stroke. HumpDay's optimisers tune aim angle, hit power, and backspin; the score is how close the ball comes to rest from the hole — plus a big bonus for actually sinking it.
Try a manual putt — slide, strike, see if your eye beats the algorithm.
Each stroke is animated at 2× the final-replay speed.
Best stroke per algorithm. Holed putts beat near-misses; the obstacle-laden course rewards algorithms that find the chip-over-the-wall path.
| Algorithm | Score | Strokes used | Best params (angle° / power / backspin) |
|---|---|---|---|
| — no runs yet — | |||
The course is deliberately bimodal. The lower green is reachable with a low-arc, modest-power stroke but has no hole — the slope bleeds the ball forward into the gap, so this basin caps at ~30-45 score. The upper green is steeper, higher, and farther; reaching it needs the arc to clear both the wall and the balloon, the power to make it that far, and enough backspin to brake on the downhill surface. The balloon prevents a degenerate strategy of just lobbing everything sky-high.
Local searchers (PRIMA_BOBYQA, Nelder-Mead) tend to settle on the lower-green basin once they find any non-zero score there. Population-based methods (CMA-ES, DE) sample broadly enough to discover the upper-green basin and then refine inside it. The gap between those two outcomes is what makes the leaderboard interesting.
Physics powered by Matter.js.
If your hyper-parameter searches are heating the Earth, drop this in Cursor or Claude:
Read https://raw.githubusercontent.com/microprediction/humpday/main/SKILL.md and create a project skill from it.