Every demo on this site is one problem. Here they are all at once: a live tournament where eleven HumpDay optimizers each get the same small budget on fourteen very different objectives — three textbook test functions and eleven drawn straight from the demos. Each cell is a rank (1 = best on that problem). Press the button and watch it fill in.
The point is what you won't see: a column of all-1s. There is no optimizer that wins everywhere. The one that looks best on average is always near-worst on something — that's the No Free Lunch theorem, made of measurements rather than maths.
The three classic functions — Rosenbrock (a smooth banana-shaped valley), Rastrigin and Ackley (egg-carton landscapes riddled with local minima) — are the standard yardsticks. The other eleven are the real thing, lifted straight from the demos: the rugged-needle focus of Lens Design, the noisy Tetris and Plinko scores, the emergent gait of the Walking Creature, the 60-knob Genetic Art image fit, plus Antenna gain, Ebola control, Robot Arm reach, Brachistochrone descent, Rocket landing and Battery arbitrage. Same eleven optimizers, same budget, wildly different winners.
Watch the patterns rather than the exact numbers (which jitter from run to run): interpolation methods like PRIMA_BOBYQA tend to dominate the small, smooth, rugged problems and then fall to the bottom on the high-dimensional and emergent ones; Powell swings from hero to last; population methods like Differential Evolution earn their keep exactly where the others drown — in high dimensions. This spread is the entire reason HumpDay exists: it doesn't sell you one optimizer, it helps you pick the right one for the problem in front of you.
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.