Both colonies want the far food source on the right. Both have identical biology, identical numbers of agents, identical everything — except that the one on the right has a few small, unglamorous food sources between here and there. Watch what each one actually banks, and when.
The colony on the left cannot smell the moon. Not because it isn't trying — it has exactly as many agents working exactly as hard — but because there is no gradient to climb from where it is standing. It explores, it builds beautiful structure, and for a very long time it banks almost nothing — a trickle from the few agents that wander into the goal by luck. By the end it has roughly a third of what the other one has.
The colony on the right smells the first stepping stone almost immediately, because that one is close enough to sense. It consumes it, which forces it onward to the second, which it can now sense from where it ended up. Each step is a genuine no-brainer, each one pays for itself, and the chain carries it somewhere it could never have jumped to directly.
Neither colony is smarter. The difference is entirely in whether there was a sensible next step, and that was decided before either of them started.
The counter is cumulative food consumed — value actually banked, not distance travelled or goals reached. That's deliberate. A moonshot isn't bad because it never arrives; sometimes it does. It's bad because the feedback loop is so long that you learn nothing, bank nothing, and can't correct for years.
An honest note on the model. Given enough time and enough random exploration, the left-hand colony can stumble onto the moon — a bounded space with thousands of wandering agents will eventually cover it. So this doesn't show that moonshots are impossible, and it isn't meant to. It shows the thing that's actually true: one of these strategies is banking value from the first minute, and the other is spending everything it has on the hope of a payoff it cannot yet detect.
An earlier version of this tried to claim the moonshot colony simply fails. It didn't reliably, so that claim was dropped rather than dressed up.