A game where each cyclist is a mini neural network, and they try to figure out an optimal strategy.

I like this a lot.

Trying to understand: so there is a single layer between the inputs and the output? And the inputs are all computed…

Interesting that there is (I assume a running) average 100m gradient and a separate 1000m gradient. I suppose so brains can differentiate between brief climbs/descents and longer climbs/descents.

Wondering now about an input for "Less than 100m to finish" to allow for last second sprinting. Or "Rider approaching behind within 10m".