Running to stay in place: when the scoreboard lies

In Through the Looking-Glass, the Red Queen tells Alice: “it takes all the running you can do, to keep in the same place.” A biologist, Leigh Van Valen, borrowed the line for evolution. When two species compete, each is the other’s world. Predators get faster, so prey get faster, so predators get faster again. Everyone improves, and yet relative to each other, nobody pulls ahead. All that running, same place.

This is where our seventh line of research begins, and it comes with a nasty measurement trap that we wanted to stare at directly before anything fancy.

The trap

So far, every experiment on this site has measured a machine against a fixed task: balance this pole, escape this maze. The score means something because the task holds still. Coevolution breaks that. When you evolve two populations that compete, and you measure “how well is population A doing against population B right now,” the yardstick is moving. B is evolving too.

Here is the trap in one sentence: both sides can be getting genuinely, dramatically better, while the score between them sits at 50-50 the entire time. If you only watch that score, you conclude nothing is happening. You would be completely wrong.

The smallest possible demonstration

We did not want to argue about this. We wanted to see it, on a case so simple that the truth is impossible to miss. So the “strategy” of each player is just a single number. Bigger numbers tend to beat smaller ones (gently, not absolutely). Two populations of numbers coevolve, each trying to beat the other.

What happens is not subtle. The numbers climb, and climb, without limit, generation after generation. That climb is real progress you can read with your own eyes. And the head-to-head score between the two populations never budges from 50-50.

Why this is the whole point

The lesson is a rule we will carry through everything that comes next: to measure progress in a competition, never use the score against your current opponent. Use a fixed yardstick. Freeze an old opponent, or a hand-built reference, and measure against that. The frozen ruler does not run alongside you, so it can tell you how far you have actually come.

Why prove something so small and so obvious-sounding? Because the games we are building toward are not obvious at all: predators chasing prey, and eventually a little 2D world where creatures forage and compete and evolve on their own. In those, the truth is hidden. You cannot read progress off a single number. If our ruler is broken, we would never know. So we validated the ruler here, on the one game where we can also see the truth directly, and confirmed it agrees. Boring on purpose. A trustworthy instrument, checked on a known length before we measure the unknown.

We also did not pick this starting point ourselves. Our in-house critic, whose only job is to attack our plans, vetoed our first idea (jump straight to predators and prey) precisely because we would not have been able to tell a real result from a broken instrument. It sent us back to the numbers. That is the adversary we wrote about before, earning its keep again.

What comes next

This is rung one of a deliberately gentle climb: next the same test where the truth is no longer a single readable number, then games that can loop instead of progress, then games where one side runs away and the other gives up, and finally the predators and prey. Each a small, signed step. The destination is a living 2D world; the discipline is refusing to run before we can measure.

Read the rigorous version

Every number here is a signed, dated entry in the open corpus, including the exploratory first run we discarded and kept on the record. Follow the sources under this post, or start with the Programme 7 charter and the signed insights index.